CN102129557A - Method for identifying human face based on LDA subspace learning - Google Patents

Abstract

The invention discloses a method for identifying a human face based on local data area (LDA) subspace learning. The method is characterized by comprising the following steps of: performing pre-processing, namely rotating uprightly a human face image; calculating a gradient multi-scale local phase quantification (GMLPQ) characteristic set of the human face image; screening out a candidate characteristic subset in the GMLPQ characteristic set by using an Adaboost selector; analyzing to acquire a human face characteristic template by using an LDA subspace analyzer; and matching the human face characteristic template with a pre-built human face characteristic template base so as to acquire identity information of a person to be identified. In the method, an optimized central point acquired by learning is used as a central point of a cosine distance, so the effectiveness of metering the sample similarity by calculating a sample distance is improved, and the classification performance of the subspace method is enhanced. The method is high in environmental suitability, and identification rate and error identification rate under vague images, low resolution and various illumination conditions, and quick in calculation speed, particularly suitable for embedded products, and can be popularized and used on a large scale.

Description

A kind of face identification method based on the study of LDA subspace

Technical field

The invention belongs to face identification method.

Background technology

Face recognition technology is one of current biological identification technology of greatly developing.Face identification system mainly comprises data acquisition subsystem, people's face detection subsystem and recognition of face subsystem.It is the most key technology of recognition of face subsystem that face characteristic extracts, good face characteristic extractive technique will make the face characteristic value of extraction littler, distinguish that performance is better, can improve discrimination and reduce misclassification rate.Present already present face feature extraction method mainly contains: based on the geometric properties method, based on the subspace analysis method, based on the wavelet theory method, based on neural net method, based on hidden markov model approach, based on support vector machine method with based on the three-dimensional model method.Based on the thought of geometric properties method is to extract the relative position at the representative position (for example eyebrow, eyes, nose, face etc.) of people's face portion and relative size as feature, the shape information that is aided with facial contour again is as feature, this method is subjected to illumination easily, the influence of factor such as expresses one's feelings, blocks, and stability is not high.Based on the main thought of the face identification method of wavelet theory is that facial image can be used for representing people's face through the low-frequency image that obtains behind the wavelet transformation.Artificial neural network ANN interconnects the network system that forms by a large amount of simple processing units, self study, self-organization, association and fault-tolerant aspect have stronger ability, in learning process, extract the feature that obtains and can be used as face characteristic and discern.Method based on subspace analysis is present popular face identification method, basic thought is to project to the facial image of loose distribution in the higher dimensional space in the subspace of a low-dimensional by linearity or nonlinear transformation, it is compact more to make being distributed in of facial image hang down in the n-dimensional subspace n, more help classification, and calculate from higher-dimension and to become low-dimensional and calculate and to solve " dimension disaster " problem.The linear subspaces method has: pivot analysis PCA, svd SVD, linear discriminant analysis LDA, independent pivot analysis ICA and nonnegative matrix factor NMF etc., non-linear subspace method has: kernel principal component analysis, nuclear Fishe discriminatory analysis, manifold learning method etc.Characteristics such as the method for subspace analysis has that calculation cost is little, descriptive power is strong, classification property is strong are one of main stream approach of current recognition of face based on the linear discriminant analysis method of separability criterion.Traditional subspace analysis method all is to weigh similarity between the different samples with Euclidean distance in training process.For Euclidean distance, different center point coordinates is for not influence of the Euclidean distance between sample.And for the cosine distance, the difference of coordinate central point can significantly impact the tolerance of cosine distance.In existing method, suppose all that at test phase central point is positioned at initial point, ignored the optimization of central point and chosen.

Summary of the invention

The invention provides a kind of scheme of improving the problems referred to above, the LDA subspace learning method through the aftertreatment of improvement tolerance that is applied to recognition of face that a kind of performance is better, robustness is stronger is provided.

Technical scheme of the present invention provides a kind of face identification method based on the study of LDA subspace, and it is characterized in that: it may further comprise the steps:

1) obtain facial image, and with described facial image become a full member, pre-service such as filtering, regulation resolution;

2) facial image calculating 1) obtains the multiple dimensioned local phase of its gradient and quantizes the GMLPQ feature set;

3) use the Adaboost selector switch to 2) described in the multiple dimensioned local phase of gradient quantize the GMLPQ feature set and screen, filter out the feature that wherein has distinguishing ability and form candidate feature subset;

4) use LDA subspace analysis device to 3) described in candidate feature subset analyze, obtain a low dimensional feature vector as its face characteristic template;

5) with 4) described in the face characteristic template mate with the face characteristic template base of building in advance, obtain the identity information of people in the described facial image.

Preferably, the multiple dimensioned local phase of gradient step 2) quantizes that the GMLPQ feature set is based on the horizontal gradient image of described facial image and VG (vertical gradient) image and the set of the LPQ feature in the multiple dimensioned multifrequency territory of extracting.

Preferably, the subspace analysis of LDA described in step 4) device forms based on the cosine distance metric training of preferred center point, and all the other chordal distance computing functions are as follows:

$d_{\cos }=\frac{{(X-O_i)}^T(Y-O_i)}{\left|\right|X-O_i\left|\right|\left|\right|Y-O_i\left|\right|}$

Wherein Oi is the center point coordinate of i class sample correspondence.

Preferably, described preferred center point is to find the solution and obtain by objective function being used gradient descent method or gradient Conjugate method.

The present invention proposes to obtain the central point of preferred central point as the cosine distance through study, has strengthened by calculating the sample distance to measure the validity of sample similarity, has improved the classification performance of subspace method.The described face recognition technology of this method has stronger environmental suitability than existing other face recognition technology, under image blurring (losing Jiao, motion etc.), low resolution, various illumination condition (infrared, visible light), have discrimination and misclassification rate preferably, and computing velocity is fast, be particularly suitable for embedded product, can be in large-scale application.

Description of drawings

Fig. 1 is an algorithm principle block diagram of the present invention.

Embodiment

Below the specific embodiment of the present invention is described in further detail.

As shown in Figure 1, the LDA subspace learning method step of the improvement tolerance aftertreatment that is applied to recognition of face of the present invention is: earlier by digital camera, camera mode such as take pictures, obtain facial image as research object, carry out the facial image pre-service then successively, extract GMLPQ feature set, Adaboost selector switch, LDA subspace analysis device, carry out the face characteristic comparison at last.

Below in conjunction with the algorithm principle figure shown in the accompanying drawing 1, the embodiment of process in detail.

1. facial image is become a full member, pre-service such as filtering, regulation resolution.

2. the multiple dimensioned local phase of gradient of facial image quantizes (GMLPQ) feature set described in calculating 1..GMLPQ feature extraction principle:

The GMLPQ feature just is based on gradient image and extracts the MLPQ feature, and gradient image comprises horizontal gradient image and VG (vertical gradient) image.And the MLPQ feature is exactly the LPQ fusion feature in multiple dimensioned multifrequency territory, MLPQ feature extraction principle:

For image, suppose that image is subjected to the fuzzy influence of certain noise, can be expressed as on frequency domain:

G＝F·H

Wherein F is the Fourier transform of original image, and G is the Fourier transform of the image after bluring, and H is ambiguity function (Fourier transform of point spread function).Their amplitude and phase component satisfy respectively:

|G|＝|F|·|H|

∠G＝∠F+∠H

The postulated point spread function is centrosymmetric in the spatial domain, and then H is in real number field, and just the phase place of H is 0 or pi in frequency domain.Suppose again that at low frequency region the value of H is being for just, so F and G's has a common phase place.LPQ just is based on that this principle proposes, and it has a reasonable robustness to image blurring.MLPQ is an expansion to LPQ, adopts a plurality of different yardsticks to extract LPQ feature under the different frequency domains, thereby can portray people's face better.LPQ extracts flow process:

(1) image is divided into the fritter of a plurality of overlapping a certain size.

(2) in each fritter, carry out short time discrete Fourier transform respectively.

Extract the phase information under some characteristic frequency, it is encoded, obtain the LPQ feature.Original LPQ feature extracting method calculated amount is bigger, and the centre exists many double countings, is difficult to reach the extract real-time of feature.The present invention proposes a kind of quick LPQ feature extracting method based on integrogram.The LPQ feature coding:

(1) each phase quadrant is encoded with 2 bit, and for example 00 represents first quartile, 01 expression, second quadrant, 10 expression third quadrants, 11 expression four-quadrants.The phase encoding of 4 different frequency domains is linked together, obtain the binary string of 8 bit, convert the coding of metric 0-255 then to.

(2) original LPQ coding dimension is higher, and is vulnerable to The noise.Adopt the method for typical module, from sample, count the highest K of a frequency of occurrences LPQ sign indicating number, remaining sign indicating number is all merged in the sign indicating number, thereby reduced LPQ condition code dimension, accelerated computing velocity, guaranteed the robustness of LPQ sign indicating number simultaneously.

3. use the feature composition candidate feature subset that the multiple dimensioned local phase of gradient described in the screening 2. of Adaboost selector switch quantizes to have in (GMLPQ) feature set distinguishing ability.

The described algorithm of this method adopts Adaboost training characteristics selector switch.Adaboost is a kind of iterative algorithm, and its core concept is at the different sorter (Weak Classifier) of same training set training, then these Weak Classifiers is gathered, and constitutes a stronger final sorter (strong classifier).Its algorithm itself realizes by changing DATA DISTRIBUTION whether it is correct according to the classification of each sample among each training set, and the accuracy rate of the overall classification of last time, determines the weights of each sample.Give lower floor's sorter with the new data set of revising weights and train, will train the last fusion of the sorter that obtains at last, at every turn as last decision-making sorter.Use the adaboost sorter can get rid of some unnecessary training data features, notice is placed on above the crucial training data.

The Adaboost training process is as follows:

1. earlier obtain first Weak Classifier by study to N training sample;

2. sample and other the new data with misclassification constitutes a new N training sample together, obtains second Weak Classifier by the study to this sample;

With 1 and 2 all misclassification sample add that other new samples constitutes the training sample of another new N, obtains the 3rd Weak Classifier by the study to this sample;

4. final process promotes and obtains strong classifier.

4. use LDA subspace analysis device candidate feature subset described in is 3. analyzed, obtain a low dimensional feature vector as the face characteristic template.

Described LDA subspace is combined by the linear discrimination analysis device in a plurality of subspaces, comprising: original LDA, enhancing LDA (E-FLDA), direct LDA (D-LDA), kernel LDA (N-LDA), edge LDA (MFA).In the process of every kind of method, all select training sample at random, make distinct methods have more complementarity, improve the model generalization ability.

The summary of LDA subspace analysis method.

LDA subspace analysis method is exactly the linear discriminant analysis method, target is to extract the low dimensional feature with distinguishing ability in high-dimensional feature space, the sample that these features help to belong to same class flocks together more, belongs to inhomogeneous sample and separates more.A projection matrix W is found in mathematical description mode such as the definition of LDA objective function exactly, makes the ratio of interior divergence matrix S w of class and between class scatter matrix S b maximize.

Divergence matrix S w definition in the class:

$S_w=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{N_k}{\mathrm{\Σ}}}(X_{\mathrm{ij}}-m_i){(X_{\mathrm{ij}}-m_i)}^T$

Between class scatter matrix S b definition:

$S_b=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=i+1}{\overset{N}{\mathrm{\Σ}}}(m_i-m_j){(m_i-m_j)}^T$

The definition of LDA objective function:

$J=\frac{w^T{S}_{b}w}{w^T{S}_{w}w}$

Face identification method based on the LDA subspace is a multiclass problem concerning study, and a people is exactly a class, and people's face sample of same individual belongs to same class, and people's face sample of different people belongs to inhomogeneity.LDA subspace face identification method learns to obtain a low n-dimensional subspace n exactly, make human face characteristic point project to this low n-dimensional subspace n from high-dimensional feature space, the human face characteristic point that belongs to same individual is assembled more, and the human face characteristic point that belongs to different people separates more.For the multiclass problem concerning study, how objectively effectively to measure the similarity degree (confidence level that belongs to same classification) between sample, the classification performance of the sorter that will obtain study plays crucial effect, and different distance metric methods have evident difference to discrimination.In many field of image recognition, in recognition of face, scholars find to have higher discrimination than Euclidean distance or L1 distance usually with the cosine distance.The present invention points out that the cosine distance also has further improved space, has proposed to improve from traditional subspace learning outcome the aftertreatment learning algorithm of distance metric, improves by calculating the sample distance and measures the confidence level and the validity of sample similarity.

The present invention improves cosine distance metric method.For the cosine distance, the difference of coordinate central point can significantly impact the tolerance of cosine distance.In existing method, suppose all that at test phase central point is positioned at initial point, ignored the optimization of central point and chosen.The present invention proposes to obtain the central point of preferred central point as the cosine distance through study, strengthened by calculating the sample distance and measured the validity of sample similarity, guaranteed that similar sample assembles more, the inhomogeneity sample separates more, improve the classification performance of subspace method, effectively reduced people's face misclassification rate.

Suppose that Oi is the central point of current selection, the cosine distance calculation formula of improved preferred center point is as follows:

$d_{\cos }=\frac{{(X-O_i)}^T(Y-O_i)}{\left|\right|X-O_i\left|\right|\left|\right|Y-O_i\left|\right|}$

Shown in formula, can select the central point of Oi arbitrarily as the cosine distance calculation, comprise certainly and select the central point of class central point as the cosine distance.Therefore, find the central point of an optimal Oi, improve the classification performance of subspace method as the cosine distance.

The classification performance of subspace method is assessed by objective function, therefore need learn the Oi central point of the best by objective function.Suppose to have the c type objects, the sample number of each class is Nk, and total sample number is N, and i sample of k class is designated as Then optimization aim function in subspace is:

$J$

$=\underset{k=1}{\overset{C}{\mathrm{\Σ}}}\underset{i=1}{\overset{N_k}{\mathrm{\Σ}}}\underset{j=i+1}{\overset{N_k}{\mathrm{\Σ}}}\frac{{(x_i^k-O^k)}^T(X_j^k-O^k)}{\sqrt{{(X_i^k-O^k)}^T(X_i^k-O^k){(X_j^k-O^k)}^T(X_j^k-O^k)}}$

$-\underset{m=1}{\overset{C}{\mathrm{\Σ}}}\underset{n=m+1}{\overset{C}{\mathrm{\Σ}}}\underset{i\∈C_m}{\mathrm{\Σ}}\underset{j\∈C_n}{\mathrm{\Σ}}(\frac{{(X_i^m-O^m)}^T(X_j^n-O^m)}{\sqrt{{(X_i^m-O^m)}^T(X_i^m-O^m){(X_j^n-O^m)}^T(X_j^n-O^m)}}$

$+\frac{{(X_i^m-O^n)}^T(X_j^n-O^n)}{\sqrt{{(X_i^m-O^n)}^T(X_i^m-O^n){(X_j^n-O^n)}^T(X_j^n-O^n)}})$

If

$g_{\mathrm{ij}}^{\mathrm{mnk}}=\stackrel{f_{\mathrm{ij}}^{\mathrm{mnk}}={(X_i^m-O^k)}^T(X_j^n-O^k),}{\sqrt{{(X_i^m-O^k)}^T(X_i^m-O^k){(X_j^n-O^k)}^T(X_j^n-O^k)}}$

$J_1=\underset{k=1}{\overset{C}{\mathrm{\Σ}}}\underset{i=1}{\overset{N_k}{\mathrm{\Σ}}}\underset{j=i+1}{\overset{N_k}{\mathrm{\Σ}}}\frac{f_{\mathrm{ij}}^{\mathrm{kkk}}}{g_{\mathrm{ij}}^{\mathrm{kkk}}}$

$J_2=\underset{m=1}{\overset{C}{\mathrm{\Σ}}}\underset{n=m+1}{\overset{C}{\mathrm{\Σ}}}\underset{i\∈C_m}{\mathrm{\Σ}}\underset{j\∈C_n}{\mathrm{\Σ}}\frac{f_{\mathrm{ij}}^{\mathrm{mnm}}}{g_{\mathrm{ij}}^{\mathrm{mnm}}}$

$J_3=\underset{m=1}{\overset{C}{\mathrm{\Σ}}}\underset{n=m+1}{\overset{C}{\mathrm{\Σ}}}\underset{i\∈C_m}{\mathrm{\Σ}}\underset{j\∈C_n}{\mathrm{\Σ}}\frac{f_{\mathrm{ij}}^{\mathrm{mnn}}}{g_{\mathrm{ij}}^{\mathrm{mnn}}}$

Then

J＝J ₁-J ₂-J ₃

The methods such as gradient descent method or method of conjugate gradient of using are optimized improved LDA objective function, and each objective function J1, J2, J3 are asked local derviation.Change central point Oi, make objective function J reduce the fastest direction and walk, when determining the objective function of overall situation the best, obtain the central point of best central point O as the cosine distance towards gradient.

$\frac{{\∂J}_1}{{\∂O}^k}=\underset{k=1}{\overset{C}{\mathrm{\Σ}}}\underset{i=1}{\overset{N_k}{\mathrm{\Σ}}}\underset{j=i+1}{\overset{N_k}{\mathrm{\Σ}}}\frac{{\left(f_{\mathrm{ij}}^{\mathrm{kkk}}\right)}^{\′}\left(g_{\mathrm{ij}}^{\mathrm{kkk}}\right)-\left(f_{\mathrm{ij}}^{\mathrm{kkk}}\right){\left(g_{\mathrm{ij}}^{\mathrm{kkk}}\right)}^{\′}}{{\left(g_{\mathrm{ij}}^{\mathrm{kkk}}\right)}^2}$

$\frac{{\∂J}_2}{{\∂O}^k}=\underset{n=k+1}{\overset{C}{\mathrm{\Σ}}}\underset{i\∈C_k}{\mathrm{\Σ}}\underset{j\∈C_n}{\mathrm{\Σ}}\frac{{\left(f_{\mathrm{ij}}^{\mathrm{knk}}\right)}^{\′}\left(g_{\mathrm{ij}}^{\mathrm{knk}}\right)-\left(f_{\mathrm{ij}}^{\mathrm{knk}}\right){\left(g_{\mathrm{ij}}^{\mathrm{knk}}\right)}^{\′}}{{\left(g_{\mathrm{ij}}^{\mathrm{knk}}\right)}^2}$

$\frac{{\∂J}_3}{{\∂O}^k}=\underset{m=1}{\overset{k-1}{\mathrm{\Σ}}}\underset{i\∈C_m}{\mathrm{\Σ}}\underset{j\∈C_k}{\mathrm{\Σ}}\frac{{\left(f_{\mathrm{ij}}^{\mathrm{mkk}}\right)}^{\′}\left(g_{\mathrm{ij}}^{\mathrm{mkk}}\right)-\left(f_{\mathrm{ij}}^{\mathrm{mkk}}\right){\left(g_{\mathrm{ij}}^{\mathrm{mkk}}\right)}^{\′}}{{\left(g_{\mathrm{ij}}^{\mathrm{mkk}}\right)}^2}$

$g^{\′}=\frac{1}{2g}((X-O){(X-O)}^T(X-O)$

$+(X-O){(X-O)}^T(X-O)$

$+(X-O){(X-O)}^T(X-O)+(X$

$-O)(X-O)^T(X-O))$

${\left(f_{\mathrm{ij}}^{\mathrm{kkk}}\right)}^{\′}=-(X_j^k-O^k)-(X_i^k-O^k)$

${\left(f_{\mathrm{ij}}^{\mathrm{knk}}\right)}^{\′}=-(X_j^n-O^k)-(X_i^k-O^k)$

${\left(f_{\mathrm{ij}}^{\mathrm{knk}}\right)}^{\′}=-(X_j^k-O^k)-(X_i^m-O^k)$

${\left(g_{\mathrm{ij}}^{\mathrm{kkk}}\right)}^{\′}=\frac{1}{{2g}_{\mathrm{ij}}^{\mathrm{kkk}}}((X_i^k-O^k){(X_j^k-O^k)}^T(X_j^k$

$-O^k)$

$+(X_i^k-O^k){(X_j^k-O^k)}^T(X_j^k-O^k)$

$+(X_j^k-O^k){(X_i^k-O^k)}^T(X_i^k-O^k)$

$+(X_j^k-O^k)(X_i^k-O^k)^T(X_i^k$

$-O^k\left)\right)$

${\left(g_{\mathrm{ij}}^{\mathrm{knk}}\right)}^{\′}=\frac{1}{{2g}_{\mathrm{ij}}^{\mathrm{knk}}}((X_i^k-O^k){(X_j^n-O^k)}^T(X_j^n$

$-O^k)$

$+(X_i^k-O^k){(X_j^n-O^k)}^T(X_j^n-O^k)$

$+(X_j^n-O^k){(X_i^k-O^k)}^T(X_i^k-O^k)$

$+(X_j^n-O^k)(X_i^k-O^k)^T(X_i^k$

$-O^k\left)\right)$

${\left(g_{\mathrm{ij}}^{\mathrm{mkk}}\right)}^{\′}=\frac{1}{{2g}_{\mathrm{ij}}^{\mathrm{mkk}}}((X_i^m-O^k){(X_j^k-O^k)}^T(X_j^k$

$-O^k)$

$+(X_i^m-O^k){(X_j^k-O^k)}^T(X_j^k-O^k)$

$+(X_j^k-O^k){(X_i^m-O^k)}^T(X_i^m-O^k)$

$+(X_j^k-O^k)(X_i^m-O^k)^T(X_i^m$

$-O^k\left)\right)$

Method of testing: for C class sample, establishing Y is the input sample, belongs to the enrolled set sample X of K class, and the similarity distance calculation of Y and X is as follows:

d _cos＝(Y-O ^k) ^T(X-O ^k)

5. face characteristic template described in will be 4. with build the face characteristic template base in advance and mate, obtain identification people identity information.

More than describe the improvement that is applied to recognition of face of the present invention in detail and measure the implementation and the realization details thereof of the LDA subspace learning method of aftertreatment.

Above embodiment only is the present invention's a kind of embodiment wherein, and it describes comparatively concrete and detailed, but can not therefore be interpreted as the restriction to claim of the present invention.Should be pointed out that for the person of ordinary skill of the art without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (4)

1. face identification method based on LDA subspace study, it is characterized in that: it may further comprise the steps:

1) obtain facial image, and with described facial image become a full member, pre-service such as filtering, regulation resolution;

2) facial image calculating 1) obtains the multiple dimensioned local phase of its gradient and quantizes the GMLPQ feature set;

3) use the Adaboost selector switch to 2) described in the multiple dimensioned local phase of gradient quantize the GMLPQ feature set and screen, filter out the feature that wherein has distinguishing ability and form candidate feature subset;

4) use LDA subspace analysis device to 3) described in candidate feature subset analyze, obtain a low dimensional feature vector as its face characteristic template;

5) with 4) described in the face characteristic template mate with the face characteristic template base of building in advance, obtain the identity information of people in the described facial image.

2. a kind of face identification method based on LDA subspace study according to claim 1 is characterized in that: step 2) described in the multiple dimensioned local phase of gradient quantize that the GMLPQ feature set is based on the horizontal gradient image of described facial image and VG (vertical gradient) image and the set of the LPQ feature in the multiple dimensioned multifrequency territory of extracting.

3. a kind of face identification method based on the study of LDA subspace according to claim 1 is characterized in that: the subspace analysis of LDA described in step 4) device forms based on the cosine distance metric training of preferred center point, and all the other chordal distance computing functions are as follows:

$d_{\cos }=\frac{{(X-O_i)}^T(Y-O_i)}{\left|\right|X-O_i\left|\right|\left|\right|Y-O_i\left|\right|}$

Wherein Oi is the center point coordinate of i class sample correspondence.

4. a kind of face identification method based on LDA subspace study according to claim 3 is characterized in that: described preferred center point is to find the solution and obtain by objective function being used gradient descent method or gradient Conjugate method.

Images