CN101853397A - Bionic human face detection method based on human visual characteristics - Google Patents

Abstract

The invention discloses a bionic human face detection method based on human visual characteristics, which comprises the following steps that: step 1: an input image containing a human face is pre-processed; step 2: regional rule judgment and human eye block matching screening are carried out to the pre-processed image to obtain a screening area of human eye blocks; step 3: regional rule judgment and multi-relation template block matching are carried out to the screening area of human eye blocks, and a human face area is extracted; and step 4: the human face rules of the extracted human face area are verified to obtain the position of the human face in the image. The bionic human face detection method based on human visual characteristics greatly reduces the impact of noise and complicated background on a human face detection method based of the characteristics, overcomes the defects of the human face detection method based on statistical theory that the complicacy is high, labor is wasted in searching training samples and the generalization capability is not strong.

Description

A kind of bionic human face detection method based on human visual system

Technical field

The present invention relates to technical field of image processing, especially a kind of bionic human face detection method based on human visual system.

Background technology

It is a very important job that people's face detects (from moving face location), is many application, the committed step of for example recognition of face, the repairing of people's face, face tracking and the supervision of people's face.Early stage recognition of face research is primarily aimed at the facial image (as the image of no background) that has than the strong constraint condition, often suppose people's face position and the known or easy acquisition of size, so people's face detection problem is not subjected to enough attention.And along with the sharp increase of practical application request, a series of practical problemss that faced thus make people's face detect beginning as one independently problem be subjected to researcher's attention.

At present, human face detection tech is also very unripe, and the principal element that influences people's face testing result has: (1) illumination condition, for example, intensity variations, natural light and artificial light etc.; (2) variation of attitude; (3) variation of image size; (4) cosmetic or attachment on the face such as beard and glasses etc.; (5) background is as indoor scenery and outdoor physical environment etc.Though these factors do not constitute too big obstacle for the mankind's vision system, but existing face detection system can only be obtained quite good detecting effectiveness under certain restrictive condition, does not also reach practical level at aspects such as loss, false alarm rate, detection speed and adaptability.

The basic thought that computer face detects is with method knowledge or that add up people's face to be carried out modeling, more all possible zone to be checked and faceform's matching degree, thus obtain the human face region that may exist.Existing method for detecting human face can be divided into following a few class:

1) based on the method for detecting human face of geometric properties: the geometric properties of so-called people's face refers to the feature that human face organ embodies on how much, be human under different attitudes, visual angle and illumination condition, the feature that the parts of people's face or people's face may exist with unchangeability.Utilize this feature just can detect and whether have people's face in the image.Difference according to the use characteristic type can be divided into: a) based on the method for priori, be the method for detecting human face with the coding of the relation between people's face portion organ criterionization; B) based on the method for feature unchangeability, be conceived to detect some constant features of people's face portion, for example eyes, nose and face etc. utilize various means to seek the invariant features of above-mentioned organ, and comprehensive then these invariant features that find determine whether zone to be detected is people's face; C) based on the method for template, at first make face template, calculate the correlation of surveyed area and template then, utilize the criterion of formulating to judge whether surveyed area is people's face.Different according to template construct and parameter setting can be divided into two kinds of methods of pre-solid plate and deforming template.

Based on geometric properties method for detecting human face have intuitively, easily by people's acceptance with the advantage that adopts, designs easily at different applicable cases; But, want to find a notable feature that is applicable to all situations very difficult, or even impossible; Because the information that uses is the feature of low level in the image, as Points And lines etc., these features are subjected to illumination and The noise easily, cause flase drop easily.

2) based on the method for detecting human face of complexion model: the colour of skin is people's face surface one of notable attribute the most, have relative stability and and most of backgrounds there are differences, and do not rely on the minutia of people's face portion, size for expression, attitude and image is also insensitive, so features of skin colors is the most frequently used a kind of feature in people's face detects.Primary what solve is the detection problem of skin pixel point and detect based on people's face of colour of skin information.

Method for detecting human face based on complexion model has the application of being easy to, and is insensitive to human face posture, verification and measurement ratio advantages of higher under the environment of background dullness.But, extremely important for the selection of chrominance space, influenced the effect of whole algorithm to a great extent.Under illumination and background complicated situation, effect is not very desirable based on the method for complexion model.

3) based on the method for detecting human face of statistical theory: the basic thought of these class methods is the pattern recognition problem that people's face detection problem are considered as a broad sense, be to utilize the method for statistical study and machine learning to seek out people's face sample and non-face sample statistical nature separately, re-use feature construction sorter separately, use sorter to finish people's face and detect.Method for detecting human face based on statistical nature mainly contains: the linear subspaces method (mainly comprises pca method (PCA), linear discriminant analysis method (LDA), factor analysis method (FA) etc.), neural net method, support vector machine (SVM) method, hidden Markov model (hmm) method, Bayers decision-making technique and Boosting method etc.

Method for detecting human face based on statistical theory, do not need to extract concrete face characteristic, but from a large amount of typical data, obtain statistical nature, can reflect the difference between facial image and the non-face image preferably, so strong robustness, even higher accuracy is also arranged in the image of complex background; But, need carry out exhaustive search to full figure, the computational complexity height; In order to reach higher accuracy, need a large amount of time and efforts to collect people's face sample and non-face sample; Generalization ability is not strong.

Summary of the invention

(1) technical matters that will solve

The objective of the invention is to propose a kind of bionic human face detection method based on human visual system, to overcome the shortcoming that is subjected to noise and complex background influence based on the method for detecting human face of feature easily, and overcome method for detecting human face complexity height based on statistical theory, collection training sample effort, the shortcoming that generalization ability is not strong.

(2) technical scheme

For achieving the above object, the invention provides a kind of bionic human face detection method based on human visual system, comprising:

Step 1: the image that contains people's face to input carries out the image pre-service;

Step 2: pretreated image is carried out the screening of regional rule judgment and human eye block matching, obtain the screening area of human eye block;

Step 3: the screening area to human eye block carries out regional rule judgment and many relation template piecemeal coupling, extracts human face region;

Step 4: the human face region that extracts is carried out the verification of people's face rule, obtain the position of people's face in image.

In the such scheme, the image that contains people's face to input described in the step 1 carries out the image pre-service, is that the image of importing that contains people's face is carried out illumination homogenising and normalized, to improve the quality of image.

In the such scheme, the image that contains people's face of described input comprises the image that contains upright people's face, and the image that contains colourful attitude people's face.

In the such scheme, regional rule judgment described in the step 2 and 3, be checking between the piecemeal gray integration and the symmetry and the heterogeneite of dispersion, to guarantee to meet the blocking characteristic of human eye and people's face.

In the such scheme, described in the step 2 image being carried out the screening of regional rule judgment and human eye block matching, is to be earlier the characteristic of face understanding to be carried out according to human knowledge people's face.

In the such scheme, screening area to human eye block described in the step 3 carries out regional rule judgment and many relation template piecemeal coupling, be to be the characteristic that adapts to different sizes and different far and near object by the size of regulating pupil, take the face templates of how association to carry out multiple dimensioned search matched according to human vision.

In the such scheme, described many relation template piecemeal coupling adopts the criterion of relative coefficient as the metrics match degree.

In the such scheme, described in the step 4 human face region that extracts is carried out the verification of people's face rule, be meant the gradient rule verification of " face district " by coupling being carried out the VG (vertical gradient) image, and the verification of the two-value of typical gradient image rule, final accurately people from location face.

(3) beneficial effect

From technique scheme as can be seen, the present invention has following beneficial effect:

1, this bionic human face detection method provided by the invention based on human visual system, can either greatly reduce the influence that is subjected to noise and complex background based on the method for detecting human face of feature easily, can overcome method for detecting human face complexity height again based on statistical theory, collection training sample effort, the shortcoming that generalization ability is not strong.

2, this bionic human face detection method based on human visual system provided by the invention belongs to the category based on the method for detecting human face of geometric properties, and this method can be located the zone at people's face place exactly in the image that comprises people's face.

Description of drawings

For further specifying content of the present invention and characteristics, below in conjunction with drawings and Examples the present invention is done a detailed description, wherein:

Fig. 1 is the method flow diagram that detects based on the bionic human face of human visual system provided by the invention;

Fig. 2 is the block diagram of human eye template symmetry and heterogeneite;

Fig. 3 is the block diagram of face template symmetry and heterogeneite;

Fig. 4 is the template of human eye and multiple dimensioned people's face;

Fig. 5 is three fens graph models of people's face of adopting of regular verification and numbering; Wherein, (a) being three fens graph models of people's face, (b) is three components numberings.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.

The human visual system is to the understanding of object, is to adapt to the far and near and big or small of object by the size of regulating pupil, and be the understanding by face earlier to the understanding of people's face.This characteristic of simulating human vision system, the present invention takes the method for how related multiple dimensioned template matches in people's face detects, five kinds of face templates of prefabricated human eye and different length breadth ratios, at first candidate regions is carried out preliminary screening, detect to adapt to the different shapes of face with face template then with the human eye template; Simultaneously, when coupling, adopt the subregion piecemeal coupling of human eye and people's face.This method is applicable to the image that contains people's face, by the image pre-service, eliminates the austerity of light influence, enhancing people face pattern; Then the image after handling is carried out eye district and face candidate regions rule judgment and many relation template piecemeal coupling, after accurately locate people's face after the regular verification.

As shown in Figure 1, Fig. 1 is the method flow diagram that detects based on the bionic human face of human visual system provided by the invention, and this method comprises:

Step 1: the image that contains people's face to input carries out the image pre-service;

Step 2: pretreated image is carried out the screening of regional rule judgment and human eye block matching, obtain the screening area of human eye block;

Step 3: the screening area to human eye block carries out regional rule judgment and many relation template piecemeal coupling, extracts human face region;

Step 4: the human face region that extracts is carried out the verification of people's face rule, obtain people's face position in image.

Wherein, the image that contains people's face to input described in the step 1 carries out the image pre-service, is that the image of importing that contains people's face is carried out illumination homogenising and normalized, to improve the quality of image.The image that contains people's face of described input comprises the image that contains upright people's face, and the image that contains colourful attitude people's face.

Regional rule judgment described in the step 2 and 3, be checking between the piecemeal gray integration and the symmetry and the heterogeneite of dispersion, to guarantee to meet the blocking characteristic of human eye and people's face.Described image being carried out the screening of regional rule judgment and human eye block matching, is to be earlier the characteristic of face understanding to be carried out according to human knowledge people's face.Described screening area to human eye block carries out regional rule judgment and many relation template piecemeal coupling, be to be the characteristic that adapts to different sizes and different far and near object by the size of regulating pupil, take the face templates of how association to carry out multiple dimensioned search matched according to human vision.Described many relation template piecemeal coupling adopts the criterion of relative coefficient as the metrics match degree.

Described in the step 4 human face region that extracts is carried out the verification of people's face rule, be meant the gradient rule verification of " face district " by coupling being carried out the VG (vertical gradient) image, and the verification of the two-value of typical gradient image rule, final accurately people from location face.

Based on the method flow diagram that the bionic human face based on human visual system shown in Figure 1 detects, the present invention is described in more detail below in conjunction with specific embodiment.

Step 1: image pre-service

Because the image irradiation that a variety of causes may cause image acquisition to obtain is not enough, fuzzy, effectively the image Preprocessing Algorithm can be improved the quality of image.At first coloured image is carried out gray processing, then image is carried out the illumination gradient correction, promptly utilize original image to simulate one and proofread and correct the illumination plane, deduct this plane then.

If gray level image is of a size of W * H, total number of pixels is n=W * H, and gray scale is x _g(i _k, j _k) (k=1,2 ..., n), the illumination plane that need simulate is y=θ ₁* i+ θ ₂* j+ θ ₃, then should satisfy:

$({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,{\mathrm{\θ}}_2)=\frac{\arg }{({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,{\mathrm{\θ}}_2)}\min \left(\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x(i_k,j_k)-({\mathrm{\θ}}_1*i_k+{\mathrm{\θ}}_2*j_k+{\mathrm{\θ}}_3))}^2\right)$

Write as matrix form, asked the minimum value of following quadratic form exactly:

E(θ)＝(Aθ-X) ^T(Aθ-X)

Wherein: $A=\left[\begin{array}{ccc}{i}_1& {\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="HSA00000114250600022.png"\; state="\{\{state\}\}">j</figure-callout>}}_1& 1\\ \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;\\ i_{\mathrm{<figure-callout\; id="1"\; label="n\; j\; n"\; filenames="HSA00000114250600022.png"\; state="\{\{state\}\}">n</figure-callout>}}& j_n{}_{}& 1\end{array}\right],$ $\mathrm{\&theta;}=\left[\begin{array}{c}{\mathrm{\&theta;}}_1\\ {\mathrm{\&theta;}}_2\\ {\mathrm{\&theta;}}_3\end{array}\right],$ $X=\left[\begin{array}{c}x(i_1,j_1)\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ x(i_n,j_n)\end{array}\right].$

The revised image x of illumination then _p(i _k, j _k) be:

x _p(i _k, j _k)=x _g(i _k, j _k)-y _k=x _g(i _k, j _k)-(θ ₁* i _k+ θ ₂* j _k+ θ ₃) the illumination gradient correction eliminated the single order variable quantity of image, reduced the influence of non-front lighting to a certain extent, weakened the shade of people's face.

To the revised image of illumination, carry out the variance normalization and handle, promptly the average and the variance of image are carried out normalization, it is adjusted to designated value, can strengthen the austerity of people's face pattern.x _p(i _k, j _k) be converted into image x _p(i, j) (0≤i＜H, 0≤j＜W), establish image x _p(i, average j) and variance are respectively m and σ, and desired average and variance are respectively m ₀And σ ₀, the normalized image x that then obtains _n(i j) is expressed as follows:

$x_n(i,j)={\mathrm{<figure-callout\; id="0"\; label="m"\; filenames="HSA00000114250600022.png"\; state="\{\{state\}\}">m</figure-callout>}}_0+\frac{{\mathrm{\&sigma;}}_0}{\mathrm{\&sigma;}}(x_p(i,j)-m),0\&le;i\&le;H,0\&le;j\&le;W$

Step 2: the human eye candidate regions detects screening

Human knowledge people's face is earlier to the understanding of face, therefore earlier by the eye template scalping face candidate region of choosing, and earlier the anticipation of human eye area rule is carried out in the zone before the eye template matches.

If the gray level image of segmented areas (be as the label of Fig. 21 position) is x _b(i, j), size is w * h, its gray average and dispersion are defined as follows:

$M=\frac{1}{w\&CenterDot;h}\underset{i=0}{\overset{h-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{w-1}{\mathrm{\&Sigma;}}}{x}_b(i,j),$

$\mathrm{Var}=\frac{1}{w\&CenterDot;h}\underset{i=0}{\overset{h-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{w-1}{\mathrm{\&Sigma;}}}{x_b}^2(i,j)-{\mathrm{<figure-callout\; id="2"\; label="M"\; filenames="HSA00000114250600012.png"\; state="\{\{state\}\}">M</figure-callout>}}^2,$

The gray average in piecemeal eye district and dispersion exist certain symmetry and heterogeneite (as shown in Figure 2).

1) gray average rule judgment (k ₁, k ₂, k ₃Be the constant of setting)

M ₁/k ₁＜M ₂＜k ₁×M ₁

M ₃/k ₂＜M ₅＜k ₂×M ₃

M ₃＜k ₃×M ₄，M ₅＜k ₃×M ₄

2) dispersion rule judgment (k ₄, k ₅, k ₆Be the constant of setting)

Var ₁/k ₄＜Var ₂＜k ₄×Var ₁

Var ₃/k ₅＜Var ₅＜k ₅×Var ₃

Var ₃＞k ₆×Var ₄，Var ₅＞k ₆×Var ₄

To candidate regions by above-mentioned two regular anticipations, carry out the piecemeal coupling of a template, adopt the criterion of relative coefficient (also can be called angular distance) as the metrics match degree.

If template piecemeal number is N, every segmented areas is of a size of W _m* H _m, (m=1,2 ... N), gray matrix is

Gray average is μ _T, variance is σ _T, with template be with the gradation of image matrix of the piecemeal to be matched of yardstick

Gray average is μ _X, variance is σ _X, correlation coefficient r then _m(T X) is:

$r_m(T,X)=\frac{\underset{i=0}{\overset{H_m-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{W_m-1}{\mathrm{\&Sigma;}}}(t_{i,j}-{\mathrm{\&mu;}}_T)(x_{i,j}-{\mathrm{\&mu;}}_X)}{W_m\&CenterDot;H_m\&CenterDot;{\mathrm{\&sigma;}}_T\&CenterDot;{\mathrm{\&sigma;}}_X}$

Correlation coefficient r _m(T shows that the matching degree of template and block image to be matched is high more X) greatly more.

The matching degree r of whole surveyed area is defined as follows:

$r={\left(\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left(\frac{(1+r_m)}{2}\right)}^2\right)}^{\frac{1}{2}}$

R in the formula _mThe matching degree of representing the m piece.

Candidate regions to by above-mentioned human eye rule anticipation and matching detection sorts from high to low according to its matching degree.

Step 3: obtain people's face candidate regions

The zone that obtains with the second step screening is as the zone of detecting people's face, and with the anticipation the same (people's face anticipation piecemeal is as shown in Figure 3) of human eye rule, entering candidate's face district at first needs to satisfy following rule:

1) gray average rule judgment (f ₁, f ₂, f ₃, f ₄And f ₅Be the constant of setting)

M _a/f ₁＜M _b＜f ₁×M _b

M ₁/f ₂＜M ₃＜f ₂×M ₁

M ₁＞f ₃×M ₂，M ₃＞f ₃×M ₂

M ₁＞f ₄×M ₄，M ₃＞f ₄×M ₆

M ₈＞f ₅×M ₄，M ₈＞f ₅×M ₆

2) dispersion rule judgment (f ₆And f ₇Be the constant of setting)

Var _a/f ₆＜Var _b＜f ₆×Var _b

Var ₁/f ₇＜Var ₃＜f ₇×Var ₁

Var ₁＞Var ₂，Var ₃＞Var ₂

Var ₁＞Var ₄，Var ₃＞Var ₆

To carry out multiple dimensioned face template coupling (multiple dimensioned face template as shown in Figure 4) by the candidate regions after the above-mentioned regular anticipation, take piecemeal coupling and the metrics match degree criterion same with human eye area, to be decided to be the candidate regions matching degree with the maximum match degree of multiple dimensioned template, the zone that will meet certain matching degree is as people's face candidate regions.

Step 4: regular verification

Template matches is the coupling of the overall situation just, and is not local coupling, can not show the characteristic of people's face part, and the people's face that only depends on template matches to obtain is also unreliable, also needs the further affirmation of face characteristic rule.At first to people's face candidate regions to be confirmed, use VG (vertical gradient) operator and typical gradient operator to obtain VG (vertical gradient) image and typical gradient image, specific algorithm is as follows:

If the gray level image of people's face candidate regions is [f (i, j)], the VG (vertical gradient) image is [V (i, j)], and typical gradient image is [D (i, j)], then has:

[V(i，j)]＝|f(i，j)-f(i+1，j)|；

[D(i，j)]＝|f(i，j)-f(i+1，j)|+|f(i，j)-f(i，j+1)|。

Utilize its average to carry out binaryzation to typical gradient image [D (i, j)], obtain bianry image [B (i, j)].Will [V (i, j)] and [B (i, j)] be divided into the identical block (as shown in Figure 5) of 3 * 3 sizes, add up the gray-scale value of each piecemeal.If candidate's face district is " people's face ", then should satisfy following rule:

(1) the gradient rule (

The average gray value of the sub-piece of k in expression VG (vertical gradient) image [V (i, j)] three components)

Rule?1： $\stackrel{\&OverBar;}{V}\left[0\right]>\stackrel{\&OverBar;}{V}\left[1\right],$ And $\stackrel{\&OverBar;}{V}\left[0\right]>\stackrel{\&OverBar;}{V}\left[3\right],$ And $\stackrel{\&OverBar;}{V}\left[0\right]>\stackrel{\&OverBar;}{V}\left[6\right]$

Rule?2： $\stackrel{\&OverBar;}{V}\left[2\right]>\stackrel{\&OverBar;}{V}\left[1\right],$ And $\stackrel{\&OverBar;}{V}\left[2\right]>\stackrel{\&OverBar;}{V}\left[5\right],$ And $\stackrel{\&OverBar;}{V}\left[2\right]>\stackrel{\&OverBar;}{V}\left[8\right]$

Rule?3： $\stackrel{\&OverBar;}{V}\left[6\right]<1.5\stackrel{\&OverBar;}{V}\left[7\right],$ And $\stackrel{\&OverBar;}{V}\left[8\right]<1.5\stackrel{\&OverBar;}{V}\left[7\right]$

Rule?4： $\left(\stackrel{\&OverBar;}{V}\right[7]+0.5\stackrel{\&OverBar;}{V}[6]+0.5\stackrel{\&OverBar;}{V}[8\left]\right)>1.5\stackrel{\&OverBar;}{V}\left[3\right],$

And

(2) the two-value rule ( In the presentation video [B (i, j)] in the sub-piece of k gray scale be the shared piece ratio of 1 pixel)

Rule?1： $\stackrel{\&OverBar;}{B}\left[0\right]>0.3,$ And $\stackrel{\&OverBar;}{B}\left[0\right]>\stackrel{\&OverBar;}{B}\left[1\right],$ And $\stackrel{\&OverBar;}{B}\left[0\right]>\stackrel{\&OverBar;}{B}\left[3\right]$

Rule?2： $\stackrel{\&OverBar;}{B}\left[2\right]>0.3,$ And $\stackrel{\&OverBar;}{B}\left[2\right]>\stackrel{\&OverBar;}{B}\left[1\right],$ And $\stackrel{\&OverBar;}{B}\left[2\right]>\stackrel{\&OverBar;}{B}\left[5\right]$

Rule?3： $\stackrel{\&OverBar;}{B}\left[4\right]>0$

Rule?4： $\left(\stackrel{\&OverBar;}{B}\right[7]+0.5\stackrel{\&OverBar;}{B}[6]+0.5\stackrel{\&OverBar;}{B}[8\left]\right)>1.5\stackrel{\&OverBar;}{B}\left[3\right],$

And $\left(\stackrel{\&OverBar;}{B}\right[7]+0.5\stackrel{\&OverBar;}{B}[6]+0.5\stackrel{\&OverBar;}{B}[8\left]\right)>1.5\stackrel{\&OverBar;}{B}\left[5\right]$

Experiment showed, that method of the present invention is concisely effective, the robustness height detects applicable to the people's face with complex background image.

Case study on implementation

Adopt the present invention that the image that some contain people's face is detected, after through the anticipation of human eye candidate regions rule, the human eye template being divided into 2 * 5 fritter mates, 200 candidate regions have been screened from big to small by matching degree, it as people's face candidate regions, after the anticipation of remarkable face candidate regions rule, is divided into multiple dimensioned face template 5 * 5 fritter and carries out the piecemeal coupling, total matching degree threshold value is made as 0.70, will carry out regular verification as people's face candidate regions greater than 0.70 candidate regions.Experiment showed, that this method can correctly detect people's face, respond well, practical.

Above-described specific embodiment; purpose of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the above only is specific embodiments of the invention; be not limited to the present invention; within the spirit and principles in the present invention all, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (8)

1. the bionic human face detection method based on human visual system is characterized in that, comprising:

Step 1: the image that contains people's face to input carries out the image pre-service;

Step 2: pretreated image is carried out the screening of regional rule judgment and human eye block matching, obtain the screening area of human eye block;

Step 3: the screening area to human eye block carries out regional rule judgment and many relation template piecemeal coupling, extracts human face region;

Step 4: the human face region that extracts is carried out the verification of people's face rule, obtain people's face position in image.

2. the bionic human face detection method based on human visual system according to claim 1, it is characterized in that, the image that contains people's face to input described in the step 1 carries out the image pre-service, be that the image of importing that contains people's face is carried out illumination homogenising and normalized, to improve the quality of image.

3. the bionic human face detection method based on human visual system according to claim 2 is characterized in that, the image that contains people's face of described input comprises the image that contains upright people's face, and the image that contains colourful attitude people's face.

4. the bionic human face detection method based on human visual system according to claim 1, it is characterized in that, regional rule judgment described in the step 2 and 3, be checking between the piecemeal gray integration and the symmetry and the heterogeneite of dispersion, to guarantee to meet the blocking characteristic of human eye and people's face.

5. the bionic human face detection method based on human visual system according to claim 1, it is characterized in that, described in the step 2 image being carried out the screening of regional rule judgment and human eye block matching, is to be earlier the characteristic of face understanding to be carried out according to human knowledge people's face.

6. the bionic human face detection method based on human visual system according to claim 1, it is characterized in that, screening area to human eye block described in the step 3 carries out regional rule judgment and many relation template piecemeal coupling, be to be the characteristic that adapts to different sizes and different far and near object by the size of regulating pupil, take the face templates of how association to carry out multiple dimensioned search matched according to human vision.

7. the bionic human face detection method based on human visual system according to claim 6 is characterized in that, described many relation template piecemeal coupling adopts the criterion of relative coefficient as the metrics match degree.

8. the bionic human face detection method based on human visual system according to claim 1, it is characterized in that, described in the step 4 human face region that extracts is carried out the verification of people's face rule, be meant the gradient rule verification of " face district " by coupling being carried out the VG (vertical gradient) image, and the verification of the two-value of typical gradient image rule, final accurately people from location face.

Images