CN1959702A - Method for positioning feature points of human face in human face recognition system - Google Patents

Abstract

A method for positioning character point of human face in human face identification system includes defining and positioning character point of human face, picking up character vector of character point on human face, utilizing a statistic model and applying statistic inference to label character point of human face so as to confirm position of required character point on human face.

Description

Man face characteristic point positioning method in the face identification system

Technical field

The present invention relates to a kind of localization method, specifically the man face characteristic point positioning method in the face identification system.

Background technology

In the face identification system, the face characteristic point location is an important step.The example in face characteristic zone comprises people's face face, as eye, and nose etc.Human face characteristic point includes but not limited to the central point that these are regional.The geometric projection localization method is a class classical way of face characteristic point location.It utilizes the otherness of face feature gray-scale value and skin gray-scale value, count gray-scale value and/or gray scale functional value and (projection) on level or the vertical direction, find out specific change point, then the change point position on the different directions is combined, thereby obtain the position of face feature point according to priori.But a defective with geometric projection localization method location human face characteristic point is the gray-scale value that it directly utilizes image, thereby is subjected to the influence of illumination easily.In addition, the judgement of each face feature point (as eye, mouth) etc. is separate, does not utilize the relativeness information of each face feature point like this.Another kind of man face characteristic point positioning method is active shape model (ASM) method, and it is proposed by Cootes.Active shape model (ASM) is a kind of parametrization shape description model, the shape that it comes description object with one group of discrete reference mark, and set up the variation model at each reference mark with PCA (principal component analysis) method, reference position and changing pattern to the reference mark retrain, thereby the whole deformation of assurance model is all the time within acceptable scope.Active shape model needs to try to achieve optimization solution with the mode that iterates, so the complicacy of algorithm can be than higher.In addition, it can only obtain locally optimal solution, but can't guarantee globally optimal solution.

Summary of the invention

The objective of the invention is to propose a kind of susceptibility that reduces illumination, considered the mutual relationship on the geometric position between each human face characteristic point, and can obtain globally optimal solution, thereby the man face characteristic point positioning method in the face identification system of raising bearing accuracy.

Purpose of the present invention can be achieved through the following technical solutions:

Man face characteristic point positioning method in the face identification system utilizes the statistical model of image gradient directional information, determines human face characteristic point by the method for statistical reasoning, may further comprise the steps:

(1) definition and location human face characteristic point promptly utilize the direction definition of image gradient and location candidate's human face characteristic point;

(2) proper vector of human face characteristic point in the extraction step (1);

(3) utilize a statistical model of having considered the feature and the relativeness of human face characteristic point, adopt the method for statistical reasoning, mark human face characteristic point, thereby the position of definite human face characteristic point that needs.

Purpose of the present invention can also further realize by following technical measures:

Man face characteristic point positioning method in the aforesaid face identification system, wherein said human face characteristic point comprise the central point and the marginal point of people's face face, the perhaps central point of other people face characteristic portion and marginal point.

Man face characteristic point positioning method in the aforesaid face identification system, wherein said step (1) is utilized the directional information definition human face characteristic point of gradient image, the face characteristic point location is realized in the projection maximum of specific direction or the mode of minimum value by searching specific gradient direction information, be may further comprise the steps:

1. to input imagery I, calculate its gradient Ix and Iy in X and Y direction;

2. Ix and Iy are carried out smothing filtering;

3. according to Ix and Iy after level and smooth, the directional image Ang of compute gradient;

4. the directional image Ang of gradient is carried out binaryzation and obtains Angmap, if promptly Ang (j, i) in a certain scope, then Angmap (j, i)=1, otherwise Angmap (j, i)=0;

5. extract the distributed intelligence of 0 and 1 pixel among the Angmap, this message reflection the positional information of human face characteristic point.

Man face characteristic point positioning method in the aforesaid face identification system, definition and location human face characteristic point may further comprise the steps:

1. in image, determine the sweep trace of some different angles;

2. to each pixel in the every sweep trace, get a sweep trace vertical with this sweep trace, in Angmap, calculate the number that near continuous 0 this pixel is worth or 1 value is put in this sweep trace, thereby to every sweep trace, obtain two one-dimensional vector, each vector value is represented the number of continuous 0 value or 1 value point;

3. this one-dimensional vector is carried out smothing filtering;

4. extract local minimum and maximum value point in the one-dimensional vector after level and smooth, these minimum and maximum value points have promptly been represented the position of candidate's human face characteristic point.

Man face characteristic point positioning method in the aforesaid face identification system, wherein said statistical model is latent Marko husband statistical model or condition random field statistical model.

Man face characteristic point positioning method in the aforesaid face identification system, utilize latent Marko husband statistical model that human face characteristic point is marked and may further comprise the steps:

1. set up the latent Marko husband statistical model of a two dimension, comprise aforesaid human face characteristic point as the unique point that observes, in this model, represent the sequence of an observed reading with X, represent a mark sequence with L, the joint probability of X and L is:

$P(X,L)=P\left(L_{Z1}\right)P\left({\mathrm{<figure-callout\; id="1"\; label="X\; Z"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">X</figure-callout>}}_Z\right|L_{Z1})P\left({\mathrm{<figure-callout\; id="1"\; label="L\; L"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">L</figure-callout>}}_L\right|L_{Z1})P\left({\mathrm{<figure-callout\; id="1"\; label="L\; R"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">L</figure-callout>}}_R\right|L_{Z1})\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Li}}\right|L_{\mathrm{Li}})P\left(L_{\mathrm{Li}}\right|L_{\mathrm{Li}-1})\&times;$

$\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Zi}}\right|L_{\mathrm{Zi}})P\left(L_{\mathrm{Zi}}\right|L_{\mathrm{Zi}-1})\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Ri}}\right|L_{\mathrm{Ri}})P\left(L_{\mathrm{Ri}}\right|L_{\mathrm{Ri}-1}),$

In the formula: X _LiExpression Li observed reading constantly, L _LiRepresent corresponding state, the L in the subscript, Z, R represent the left, middle and right sweep trace respectively, P (L _Zi| L _Zi-1) represent by state L _Zi-1To state L _ZiTransition probability, P (X _Zi| L _Zi) the expression output probability, promptly at state L _ZiFind X _ZiProbability, X _ZiBe the proper vector of extracting at this candidate feature point;

2. seek mark sequence L and make P (X|L) maximum, promptly find argmax _LP (X, L).2. wherein said step can realize by dynamic programming or exhaustive method.

The latent Marko husband statistical model of two dimension comprises the image feature information and the unique point state transitions information of each unique point.

Man face characteristic point positioning method in the aforesaid face identification system, utilize the condition random field statistical model that human face characteristic point is marked and may further comprise the steps:

1. set up a condition random field statistical model, aforesaid human face characteristic point is as the unique point that observes, and in this model, represents the stochastic variable that observes, L with X _LiDeng the corresponding state of expression, the probability of y under the condition of given x $P\left(y\right|x)=\frac{e^{\mathrm{\&Psi;}(y,x)}}{{\mathrm{\&Sigma;}}_{y^{\&prime;}}{e}^{\mathrm{\&Psi;}(y^{\&prime;},x)}},$ Wherein Ψ (y x) is potential energy function,

$\mathrm{\&Psi;}(y,x)=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{l}{\mathrm{\&Sigma;}}{f}_l^1(j,y_j,x){\mathrm{\&theta;}}_{\mathrm{<figure-callout\; id="1"\; label="l"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">l</figure-callout>}}^1+\underset{(j,k)\&Element;E}{\mathrm{\&Sigma;}}\underset{l}{\mathrm{\&Sigma;}}{f}_l^2(j,k,y_j,y_k,x){\mathrm{\&theta;}}_l^2,$ F wherein _l ¹And f _l ²Be fundamental function, θ _l ¹And θ _l ²Be parameter, E represents a figure who comprises set of node and arc collection, and (j k) has arc connected node y in the ∈ E presentation graphs _jAnd y _k

2. obtain optimum mark sequences y * and make P (y|x) maximization, i.e. y ^*=argmax _yP (y|x).Wherein 2. step can realize by the method for dynamic programming.

Two-dimensional condition random field statistical model comprises the image feature information and the unique point state transitions information of each unique point.

Advantage of the present invention is: the present invention utilizes the direction of image gradient to define and locate human face characteristic point, external change such as illumination can influence the absolute intensity value of image, what but gradient was considered is the variation of the relative gray-scale value between pixel, the directional information of gradient image is subjected to the influence of illumination just very little, thereby has reduced the susceptibility to illumination.In addition, the present invention has considered the mutual relationship on the geometric position between each human face characteristic point, by the method for statistical reasoning, can obtain globally optimal solution.The present invention has also introduced condition random field (Conditional Random Field) in the face characteristic point location, has overcome mark prejudice problem, thereby has had more performance.Comprehensive above each point, the present invention can obtain human face characteristic point locating effect more accurately.

Description of drawings

Fig. 1 is a process flow diagram of the present invention.

Fig. 2 is a human face characteristic point sweep trace synoptic diagram.

Fig. 3 is the latent Marko husband statistical model figure of two dimension.

Fig. 4 is condition random field statistical model figure.

Embodiment

The present invention proposes the localization method of the human face characteristic point in a kind of face identification system, human face characteristic point comprises the central point and the marginal point of people's face face, the perhaps central point of other people face characteristic portion and marginal point.The present invention utilizes the statistical model of image gradient directional information, determines human face characteristic point by the method for statistical reasoning, and its flow process may further comprise the steps as shown in Figure 1:

(1) definition and location human face characteristic point promptly utilize the direction definition of image gradient and location candidate's human face characteristic point;

(2) proper vector of human face characteristic point in the extraction step (1);

(3) utilize a statistical model of having considered the feature and the relativeness of human face characteristic point, adopt the method for statistical reasoning, mark human face characteristic point, thereby the position of definite human face characteristic point that needs.

This method supposition people face detects to be finished, and people's face positional information obtains.At detected human face region, scan image at first obtains the position of possible human face characteristic point.For apparent in view human face characteristic point such as eye, nose utilizes a sorter in different positions and yardstick scan image and judge whether current region comprises human face characteristic point to be detected.The realization example of sorter comprises the AdaBoost algorithm, support vector machine (SVM) etc.But can reduce many to all the other distant human face characteristic point preceding method validity.In addition, preceding method all needs independent and more loaded down with trivial details training process to everyone face characteristic point.The present invention utilizes the directional information of gradient image to define some new human face characteristic points.These unique points can realize in the projection maximum of specific direction or the mode of minimum value by searching gradient direction information.An obvious benefit utilizing the directional information of gradient image is to be subjected to the influence of external change such as illumination little.External change such as illumination can influence the absolute intensity value of image, but the gradient consideration is the variation of the relative gray-scale value between pixel, and the angle information of gradient image is subjected to the influence of illumination just very little.Can also adopt smothing filtering in addition, methods such as binaryzation further reduce the influence that is subjected to external change such as illumination.

Definition and location human face characteristic point may further comprise the steps:

1. to input imagery I, calculate its gradient Ix and Iy in X and Y direction;

2. Ix and Iy are carried out smothing filtering;

3. according to Ix and Iy after level and smooth, the directional image Ang of compute gradient;

4. the directional image Ang of gradient is carried out binaryzation and obtains Angmap, if promptly Ang (j, i) in a certain scope, then Angmap (j, i)=1, otherwise Angmap (j, i)=0;

5. extract the distributed intelligence of 0 and 1 pixel among the Angmap, this message reflection the positional information of human face characteristic point.

Can determine several sweep traces at the human face characteristic point position, as shown in Figure 2, then define and locate human face characteristic point specific implementation as follows:

1. in image, determine the sweep trace of some different angles;

2. to each pixel in the every sweep trace, get a sweep trace vertical with this sweep trace, in Angmap, calculate the number that near continuous 0 this pixel is worth or 1 value is put in this sweep trace, thereby to every sweep trace, obtain two one-dimensional vector, each vector value is represented the number of continuous 0 value or 1 value point;

3. this one-dimensional vector is carried out smothing filtering;

4. extract local minimum and maximum value point in the one-dimensional vector after level and smooth, these minimum and maximum value points have promptly been represented the position of candidate's human face characteristic point.

Obtain then can these human face characteristic points to be marked after candidate's the human face characteristic point, can realize by following two kinds of embodiments by the method for statistical reasoning:

Embodiment one

The latent Marko husband statistical model of present embodiment utilization marks human face characteristic point.Set up a faceform that following unique point is arranged earlier:

Along sweep trace 1: in the eye, in the middle of eye and the nose, in the nose, in the middle of nose and the mouth, cheek;

Along in 2: two of the sweep traces, nose top, in the nose, mouth top, in the mouth, mouth bottom, chin;

Along sweep trace 3: in the eye, in the middle of eye and the nose, in the nose, in the middle of nose and the mouth, cheek;

In addition, also have a dummy status to represent not belong to above any unique point.The state of above-mentioned unique point as markov model, we set up the latent markov model of a two dimension, as shown in Figure 3, and X wherein _LiExpression Li observed reading constantly, L _LiRepresent corresponding state, the L in the subscript, Z, R represent the left, center, right sweep trace respectively.

Represent the sequence of an observed reading with X, represent the sequence of a mark with L, the joint probability of X and L is so:

P(X，L)＝P(X _L1...X _LMX _Z1...X _ZNX _R1...X _RML _L1...L _LML _Z1...L _ZNL _R1...L _RM)

Consider the latent markov model of Fig. 3, the joint probability of X and L is approximate to be reduced to

$P(X,L)=P\left(L_{Z1}\right)P\left({\mathrm{<figure-callout\; id="1"\; label="X\; Z"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">X</figure-callout>}}_Z\right|L_{Z1})P\left({\mathrm{<figure-callout\; id="1"\; label="L\; L"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">L</figure-callout>}}_L\right|L_{Z1})P\left({\mathrm{<figure-callout\; id="1"\; label="L\; R"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">L</figure-callout>}}_R\right|L_{Z1})\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Li}}\right|L_{\mathrm{Li}})P\left(L_{\mathrm{Li}}\right|L_{\mathrm{Li}-1})\&times;$

$\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Zi}}\right|L_{\mathrm{Zi}})P\left(L_{\mathrm{Zi}}\right|L_{\mathrm{Zi}-1})\underset{i}{\mathrm{\&Pi;}}P\left(X_{\mathrm{Ri}}\right|L_{\mathrm{Ri}})P\left(L_{\mathrm{Ri}}\right|L_{\mathrm{Ri}-1})$

P (L wherein _Zi| L _Zi-1) represent by state L _Zi-1To state L _ZiTransition probability, P (X _Zi| L _Zi) the expression output probability, promptly at state L _ZiFind X _ZiProbability.X _ZiBe the proper vector of extracting at this candidate feature point.Proper vector is included on every side and extracts gray-scale value and carry out principal component analysis (PCA) in the KxK zone, also comprises the geometric position of candidate feature point.

If known candidate face unique point, the task of recognition feature point mark these points exactly.Method makes P (X|L) maximum for seeking mark sequence L, promptly finds argmax _LP (X, L).This can find the solution by the method for dynamic programming (Dynamic Programming).Dynamic programming method for solving to latent markov model commonly used also is called the Viterbi method, that we adopt is Li, the article Image Classification by aTwo-dimensional Hidden Markov Model.IEEE Trans on SignalProcessing of Najmi and Gray, Vol.48, No.2, the Viterbi method in 2000.This method has a detailed description in article, does not repeat them here.Optimal L also can be found the solution by exhaustive method.

Embodiment two

Present embodiment utilizes condition random field (Conditional Random Field) statistical model that human face characteristic point is marked, and sets up a faceform that following unique point is arranged:

Along sweep trace 1: in the eye, in the middle of eye and the nose, in the nose, in the middle of nose and the mouth, cheek;

Along in 2: two of the sweep traces, nose top, in the nose, mouth top, in the mouth, mouth bottom, chin;

Along sweep trace 3: in the eye, in the middle of eye and the nose, in the nose, in the middle of nose and the mouth, cheek;

In addition, also have a dummy status to represent not belong to above any unique point.

Along three scan-line directions, respectively choose most probable 7 candidate feature points, set up condition random field (Conditional Random Field) statistical model as shown in Figure 4.In this model, represent the stochastic variable that observes, L with X _LiDeng the corresponding state of expression.The probability of y is P (y|x) under the condition of given x, and optimum mark sequences y * makes P (y|x) maximization, i.e. y ^*=argmax _yP (y|x).

Based on condition random field (Conditional Random Field) model, can represent like this $P\left(y\right|x)=\frac{e^{\mathrm{\&Psi;}(y,x)}}{{\mathrm{\&Sigma;}}_{y^{\&prime;}}{e}^{\mathrm{\&Psi;}(y^{\&prime;},x)}},$ Wherein (y x) is potential energy function (Potential Function) to Ψ

It is defined as follows: $\mathrm{\&Psi;}(y,x)=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{l}{\mathrm{\&Sigma;}}{f}_l^1(j,y_j,x){\mathrm{\&theta;}}_{\mathrm{<figure-callout\; id="1"\; label="l"\; filenames="A20061009670900152.png"\; state="\{\{state\}\}">l</figure-callout>}}^1+\underset{(j,k)\&Element;E}{\mathrm{\&Sigma;}}\underset{l}{\mathrm{\&Sigma;}}{f}_l^2(j,k,y_j,y_k,x){\mathrm{\&theta;}}_l^2,$

F wherein _l ¹And f _l ²Be fundamental function, θ _l ¹And θ _l ²Be parameter, represent a figure who comprises set of node and arc collection, and (j k) have arc connected node y in the ∈ E presentation graphs _jAnd y _kf _l ¹And f _l ²Comprise the geometric position information of image characteristic point and the image feature information around the image characteristic point.Image feature information can comprise gray scale, color and gradient etc.Optimum y* finds the solution the method that can use dynamic programming (Dynamic Programming).Dynamic programming method for solving commonly used also is called the Viterbi method, that we adopt is Li, the article ImageClassification by a Two-dimensional Hidden Markov Model.IEEETrans on Signal Processing of Najmi and Gray, Vol.48, No.2, the Viterbi method in 2000.This method has a detailed description in article, does not repeat them here.

After obtaining human face characteristic point according to above step, meticulousr human face characteristic point can obtain by near the method for Local Search these points, and perhaps the method by the predetermined cell sampling obtains.

The present invention can also have other embodiment, and the technical scheme that equal replacement of all employings or equivalent transformation form all drops within the claimed scope of the utility model.

Claims (11)

1. the man face characteristic point positioning method in the face identification system is characterized in that: utilize the statistical model of image gradient directional information, determine human face characteristic point by the method for statistical reasoning, may further comprise the steps:

(1) definition and location human face characteristic point promptly utilize the direction definition of image gradient and location candidate's human face characteristic point;

(2) proper vector of human face characteristic point in the extraction step (1);

(3) utilize one to consider the feature of human face characteristic point and the statistical model of relativeness, adopt the method for statistical reasoning, mark human face characteristic point, thereby the position of definite human face characteristic point that needs.

2. the man face characteristic point positioning method in the face identification system as claimed in claim 1 is characterized in that: described human face characteristic point comprises the central point and the marginal point of people's face face, the perhaps central point of other people face characteristic portion and marginal point.

3. the man face characteristic point positioning method in the face identification system as claimed in claim 1, it is characterized in that: described step (1) is utilized the angle information definition human face characteristic point of gradient image, the face characteristic point location is realized in the projection maximum of specific direction or the mode of minimum value by searching specific gradient angle information, be may further comprise the steps:

1. to input imagery, calculate its gradient in level and vertical direction;

2. the gradient of level and vertical direction is carried out smothing filtering;

3. according to the gradient image after level and smooth, the directional image of compute gradient;

4. the directional image of gradient is carried out binaryzation;

5. extract in the gradient direction image after the binaryzation binarized pixel point in the distributed intelligence of spatial domain, this message reflection the positional information of human face characteristic point.

4. as the man face characteristic point positioning method in claim 1 or the 3 described face identification systems, it is characterized in that: definition and location human face characteristic point may further comprise the steps:

1. in image, determine the sweep trace of some different angles;

2. to each pixel in the every sweep trace, get a sweep trace vertical with this sweep trace, calculate the number that near continuous 0 this pixel is worth or 1 value is put in this sweep trace in the gradient direction image after binaryzation, thereby to every sweep trace, obtain two one-dimensional vector, each vector value is represented the number of continuous 0 value or 1 value point;

3. this one-dimensional vector is carried out smothing filtering;

4. extract local minimum and maximum value point in the one-dimensional vector after level and smooth, these minimum and maximum value points have promptly been represented the position of candidate's human face characteristic point.

5. the man face characteristic point positioning method in the face identification system as claimed in claim 1 is characterized in that: described statistical model is latent Marko husband statistical model or condition random field statistical model.

6. as the man face characteristic point positioning method in claim 1 or the 5 described face identification systems, it is characterized in that: utilize latent Marko husband statistical model that human face characteristic point is marked and may further comprise the steps:

1. set up the latent Marko husband statistical model of a two dimension, comprise that aforesaid human face characteristic point is as the unique point that observes;

2. seek its probability maximization of mark sequence for the observational characteristic point sequence.

7. the man face characteristic point positioning method in the face identification system as claimed in claim 6 is characterized in that: 2. described step can realize by dynamic programming or exhaustive method.

8. the man face characteristic point positioning method in the face identification system as claimed in claim 6 is characterized in that: the latent Marko husband statistical model of described two dimension comprises the image feature information and the unique point state transitions information of each unique point.

9. as the man face characteristic point positioning method in claim 1 or the 5 described face identification systems, it is characterized in that: utilize the condition random field statistical model that human face characteristic point is marked and may further comprise the steps:

1. set up a two-dimensional condition random field statistical model and comprise that aforesaid human face characteristic point is as the unique point that observes;

2. seek its probability maximization of mark sequence for the observational characteristic point sequence.

10. the man face characteristic point positioning method in the face identification system as claimed in claim 9 is characterized in that: 2. described step can realize by the method for dynamic programming.

11. the man face characteristic point positioning method in the face identification system as claimed in claim 9 is characterized in that: described two-dimensional condition random field statistical model comprises the image feature information and the unique point state transitions information of each unique point.

Images