CN106485222A - A kind of method for detecting human face being layered based on the colour of skin - Google Patents

Abstract

The present invention provides a human face detection method based on skin color layering. First, a human face skin color model is established according to human face candidate areas; The elliptical area within the range is treated as a non-face area; then the non-face area is further removed according to the texture complexity; then the face candidate area is rotated for direction normalization; finally, the face template is used for connected domain analysis. The invention utilizes the skin color characteristic of the human face to preprocess the image, thereby greatly reducing the search range of the human face. In addition, the computational complexity of the skin color feature of the face is small, and it is very robust to geometric changes such as face rotation and scaling, so using the skin color feature of the face is very useful for face detection. The method provided by the invention is fast, accurate and robust, and has a wide range of applications, and can be applied in image recognition, speech recognition, data mining, machine vision and the like.

Description

A face detection method based on skin color layering

technical field

The invention relates to a face detection method, in particular to a face detection method based on skin color layering, and belongs to the technical field of face recognition.

Background technique

Face detection is the first step in an automatic face recognition system. For a given static image or video image sequence, it is detected whether there is a human face in it. If so, it is segmented from the background and its position in the image is determined. position and size. In some occasions, the conditions for taking images can be controlled. For example, when the police take pictures of criminals, they are asked to bring a certain part of the face close to the scale. At this time, the positioning of the face is very simple. In other cases, the position of the face in the image is unknown in advance, such as photos taken in some complex backgrounds. At this time, the detection of the face will be affected by the following factors: (1) The position of the face in the image The position, rotation angle and scale are not fixed; (2) hairstyle and makeup will cover some features; (3) noise appearing in the image; the application field of face detection is mainly face information processing (verification, recognition, expression analysis, etc. ) system, video conferencing or distance education system, monitoring system and tracking, content-based image and video retrieval, etc.

Face detection methods can be roughly divided into two categories: methods based on facial features and methods based on image content. In addition, color and face motion information can be used as preprocessing for face detection. Based on the prior knowledge of the face, the feature-based method uses the underlying features of the face such as face contour, face edge, organ characteristics, template features, etc. to detect the face; while the image content-based method regards the face area as It is a two-dimensional pixel matrix, which is divided into two categories: human face and non-human face, using the scheme of sample training and recognition.

Studies have found that the distribution of skin color in the color space is relatively concentrated, and the color information can separate the face from most backgrounds to a certain extent. At present, many different color space models have been proposed for different occasions. . Once the color model is determined, skin color detection can be performed first, and possible face regions can be segmented according to their similarity in chroma and spatial correlation, and the geometric characteristics or grayscale features of the region can be used to determine whether it is a human face or not. verification.

By the mid-1990s, most face detection methods relied on extracting face features for detection. For example, operators such as Soble, Marr-Hildreth, and Laplacian are used to extract edge features from face images, and the extracted edges are matched with a predefined face edge model to infer whether there is a face. Another example is that Bur first extracts facial features, such as eyes, nose, mouth, outline, etc., integrates these features, and uses statistical analysis methods to infer whether there is a human face. The active shape model is also used in face detection. The main idea is to reconstruct the deformable template of the face shape, define an energy function, and minimize the energy function by continuously adjusting the model parameters, so that the face can be detected, and so on.

Basically, the feature-based face detection method is often transformed into a search problem of facial features. For example, expert Hamouz uses Gabor filter to detect 10 facial features to infer the existence of a human face. The advantage of the detection method based on facial features is that facial features are not sensitive to image brightness, occlusion, angle, etc. In addition, the detected feature information can also be used as a subsequent face recognition module. Of course, the disadvantage lies in the complexity of this algorithm, especially the computational complexity, and it is still difficult to deal with low-resolution images and multi-face detection.

Although color information is very useful information for face detection, color information can only segment the skin color area of the face, including the face, hands and other backgrounds, reducing the range of face detection, so only color information is used It is not enough to detect the face, and other follow-up processing is required to confirm whether the face exists in the skin color area. Faces vary from person to person and are never the same. Even a pair of twins must have some differences in their faces. Although human beings can easily detect and identify a person from the face when there are great changes in expression, age or hairstyle, etc., it is very difficult to establish a system that can fully automatically perform face recognition. . It involves a lot of knowledge in pattern recognition, image processing, computer vision, physiology, psychology, and cognitive science, and is closely related to other biometric-based identification methods and computer human-computer perception interaction fields. However, compared with other human biometric identification systems such as fingerprints, retinas, iris, genes, and palm shapes, face recognition systems are more direct and friendly, and users have no psychological barriers. It can obtain some information that is difficult to obtain by other identification systems.

Contents of the invention

The technical problem to be solved by the present invention is to provide a fast, accurate and robust human face detection method based on skin color layering.

In order to solve the above-mentioned technical problems, the technical solution of the present invention is to provide a kind of face detection method based on skin color stratification, it is characterized in that: the steps are:

Step 1: Establishment of face skin color model

1.1 Color space selection

Use the YCrCb color space to build a human face skin color model;

1.2 Nonlinear transformation of YCrCb color space

The chrominance component of the YCrCb color format is not completely independent of the brightness information Y, and the clustering area of the skin color also changes nonlinearly with the brightness information Y; therefore, the nonlinear transformation of the YCrCb color format makes Y The value has as little influence on the chroma component CrCb as possible;

1.3 Ellipse fitting

Use an ellipse to approximate the skin color area, and calculate its analytical expression to obtain the face skin color model after ellipse fitting;

Step 2: Shape Authentication

Establish a hierarchical processing model for the authentication of face candidate areas, the specific method is as follows:

1) Segmentation of skin color regions using the face skin color model;

2) In order to deal with faces in different directions, an ellipse is used to approximate the shape of the face. According to the angle of the major and minor axes of the ellipse, the rotation angle of the face is analyzed, and the ratio of the major and minor axes of the face candidate ellipse is specified. The size of the major and minor axes is set at Within a certain range, the ellipse area that is not within this range will be treated as a non-face area;

Step 3: Texture Verification

Due to the difference in color between the eyes, mouth, eyebrows, facial features and face skin in the face area, the texture of the face area is more complex than other candidate areas such as hand shape and neck, and the texture complexity is calculated according to the texture complexity To further remove non-face areas;

Step 4: Orientation Normalization

For the angle between the major and minor axes of the ellipse, rotate the face candidate area to normalize the direction;

Step 5: Connected Domain Analysis

The gray scale and texture of the eyes, mouth, and eyebrows are obviously different from other areas of the face, which can be clearly seen from the detected face area;

Firstly, the detected candidate face area is grayscaled. After analyzing the face feature image, it is processed with a gradient in the Y direction. Y represents the black space, and then the face template is used for connected domain analysis. When When the grayscale sum in the corresponding connected domain exceeds a certain threshold, the candidate face area is considered to be a human face area; when it is lower than a certain threshold, the opposite is true.

Preferably, in said step 1, the specific method of the nonlinear transformation of YCrCb color space is:

Perform piecewise linear fitting on the boundary of the chroma component CrCb, and use the boundary of the chroma component CrCb to limit the skin color clustering area. The specific formula is as follows:

C' _i is the transformed chroma component C' _b , C' _r , i is the component of b and r, Denotes the average value of C _i , is the scaling value of the C′ _i component, is the scaling value of C′ _i component in the Y direction, W is the scaling value, L _Ci is the hue direction of C′ _i , H _Ci is the saturation direction of C′ _i , K _l is the offset value of L component, K _h is the offset value in the H direction, Y _max is the maximum value, and Y _min is the minimum value;

After the above-mentioned nonlinear segmental color transformation, and then project it into the Cr'-Cb' two-dimensional space, the human face skin color model is obtained.

Preferably, in the step 1, according to the human face skin color sample training estimated value in the HHI image library, in the YCrCb space:

K _l =125, K _h =188, Y _min =16, Y _max =235.

Preferably, in said step 1, the formula of ellipse fitting is:

Among them: ec _x and ec _y are the coordinates of the center of the ellipse; a, b are the major and minor axes; θ is an arbitrary angle of rotation; when an arbitrary angle is rotated, the new elliptic equation becomes formula (10); C′ _b -c _x Indicates the projection of the chromaticity space in the x direction; C′ _b -c _y indicates the projection of the chromaticity space in the y direction.

Preferably, in step 1, in formula (9) and formula (10),

C _x = 109.38, _Cy = 152.02, θ = 2.53 radians, ec _x = 1.60, ec _y = 2.41, a = 25.39, b = 14.03.

Preferably, in the step 2, when using the human face skin color model to segment the skin color area, the non-human face area is eliminated by using a median filter and a morphological processing operator.

Preferably, in the step 3, the variance is used to calculate the texture complexity, and the non-face area is further removed according to the size of the variance.

Since the distribution of the skin color of the face in the color space is relatively concentrated, the color information can separate the face from most backgrounds to a certain extent, so choose an appropriate color model and use the skin color characteristics of the face to preprocess the image , which greatly narrows the search range of faces. In addition, the skin color feature of the face has a small computational complexity and is very robust to geometric changes such as face rotation and scaling. Therefore, the use of the skin color feature of the face is useful for face detection.

The method provided by the present invention first uses the skin color characteristic of the human face as the preprocessing of the human face detection, transfers the color image from the RGB space to the improved YCrCb color space, utilizes the characteristic that the skin color of the human face is relatively concentrated on the CrCb component, and divides The skin color area of the face; then a layered face detection model is given for the authentication of the candidate face area. The invention has a wide range of applications and can be applied in image recognition, speech recognition, data mining, machine vision and the like.

Description of drawings

Figure 1 is a block diagram of a layered face detection model.

detailed description

Below in conjunction with specific embodiment, further illustrate the present invention. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

1. Color space

Before introducing the algorithm, several color spaces are briefly introduced. A color space is a model that uses numbers to represent color attributes. No color system is universal, because colors can be expressed in different models and methods. Each color space has its own color properties. When considering the problem of color quantization, the first problem that needs to be solved is to define a color space. Several different color spaces exist for different color properties.

At present, the most widely used visual color space was first developed by the CIE (Commission Internationale de I'Eclairage) International Bioelectronics Committee in 1920, and derived from it. It is not a real color in nature. The advantage is that these primary colors can be used to represent the colors we perceive numerically.

(1) RGB color space:

Defining the light with wavelengths of 700nm, 546.1nm, and 435.8nm as three primary colors constitutes the RGB (red, green, blue) color space. The space is an additive color space. Because the color is produced by the addition of light color and light color. Therefore, the secondary color combined is always brighter than the primary color. The addition of the three primary colors of red, green, and blue at the maximum intensity produces white, and the addition of red, green, and blue of the same value produces neutral gray. The lower the value, the darker the neutral gray, and vice versa. It is a color language commonly used by electronic input devices, such as monitors, scanners, and digital cameras. These devices reproduce color by emitting or absorbing light, rather than using reflected light. The RGB space is a space suitable for machine processing, not a color space based on human vision. Its main disadvantage is that it relies on the brightness values of these three primary colors. Therefore, systems based on RGB space are very sensitive to changes in brightness, shadows and non-uniformity of illumination. In addition, its color gamut is also very narrow, and some visible light colors cannot be represented by it.

(2) CMY space:

CMY (cyan, magenta, yellow) space is a subtractive color space, which is applied to printing technology, and printed matter reproduces color by reflecting light. If you subtract any of the three primary colors of RGB from white light, then you will get the complementary colors CMY of red, green and blue respectively. The colors produced with the CMY subtractive color space are not exactly the same as the colors produced with the RGB additive color space. In particular, CMY cannot represent RGB colors by simple conversion. When you go from RGB's neutral gray to CMY in printing, it's slightly reddish-purple. This phenomenon can be remedied with black Y. But adding a fourth color destroys the equal conversion from RGB to CMYK, making the color correspondence between RGB and CMYK more complicated. There is no easy way to make them one-to-one.

(3) XYZ space:

The XYZ color space defines three imaginary primary colors X, Y, Z to which the rods and cones of the human eye are most sensitive. The characteristic of this space is that all colors are represented by these primary colors. The formula for converting from RGB to XYZ is as follows:

(4) YUV space:

The YUV color space is a basic color space used in composite color video standards. The method of distinguishing several primary colors is different. The YUV space uses the brightness information Y and the UV component corresponding to the relationship between hue and saturation to represent the color space. The UV component is also called the chroma component. The relationship from RGB to YUV conversion is as follows:

(5) YCrCb space:

YCrCb space is a color space obtained by scaling YUV space and adding offsets. It is often used in image compression such as JPEG, H.261 and MPEG fields. Its relationship with RGB space is:

2. Algorithm

The distribution of the skin color of the surface face in the color space is relatively concentrated, and the color information can separate the face from most of the background to a certain extent, so choose an appropriate color model and use the skin color characteristics of the face to predict processing, which greatly reduces the search range of the face. In addition, the computational complexity of the skin color feature of the face is small, and it is very robust to geometric changes such as face rotation and scaling. Therefore, the face detection is performed using the skin color feature of the face is useful.

1. Color space selection

The present embodiment adopts the YCrCb color space, which has the following advantages:

(1) The YCrCb color format has a composition principle similar to that of human visual perception.

(2) The YCrCb color format is widely used in television display and other fields, and is also a color representation format commonly used in many video compression codes, such as MPEG, JPEG and other standards.

(3) The YCrCb color format has the advantage of separating the brightness component of the color similar to other color formats such as HIS.

(4) Compared with some other color formats such as HIS, the calculation process and spatial coordinate representation of the YCrCb color format are relatively simple.

(5) The experimental results show that the clustering characteristics of skin color are better in YCrCb color space.

2. Nonlinear transformation of YCrCb color space

Since the color space is sensitive to the illumination on the surface of the face image, the chroma component of the color space is used to construct the skin color model. Choose CrCb to build the skin color model. However, the YCrCb color format is directly obtained from the RGB color format through linear transformation, so its chroma components do not exist completely independent of the brightness information, and the clustering area of the skin color also follows the brightness information Y. vary non-linearly.

In the YCrCb color space, the skin color clustering is in the shape of a spindle with two pointed ends, that is, in the part with larger and smaller Y values, the skin color clustering area is also reduced accordingly. It can be seen that, where the Y values are different, it is not feasible to seek the clustering area of skin color in the Cr-Cb sub-plane, and the influence caused by the different Y values must be considered, that is, the nonlinear transformation of the YCrCb color format is performed, so that the Y The value has as little effect on the chroma component CrCb as possible. Here, a piecewise linear fit is made on the boundaries of the chrominance components.

Using the 4 boundaries of Cr-Cb to limit the skin color clustering area can well adapt to areas with too bright or too dark brightness, so that the robustness of the skin color model is greatly improved. The specific formula is as follows:

C _i ' is the transformed chrominance components C' _b , C' _r , i is the component of b and r, Denotes the average value of C _i , is the scaling value of the C′ _i component, is the scaling value of C′ _i component in the Y direction, W is the scaling value, L _Ci is the hue direction of C′ _i , H _Ci is the saturation direction of C′ _i , K _l is the offset value of L component, K _h is the offset value in the H direction, Y _max is the maximum value, and Y _min is the minimum value.

K _l ＝125, K _h ＝188, these data are estimated according to the human face skin color sample training in the HHI image database, in the YCrCb space, Y _min ＝16, Y _max ＝235. After such nonlinear segmental color transformation, and then projecting it into the Cr'-Cb' two-dimensional space, a practical skin color clustering model can be obtained. According to the traditional method, an ellipse can be used to approximate this skin color area, and its analytical expression can be obtained as:

Among them: ec _x and ec _y are the coordinates of the center of the ellipse; a, b are the major and minor axes; θ is an arbitrary angle of rotation; when an arbitrary angle is rotated, the new elliptic equation becomes formula (10); C′ _b -c _x Indicates the projection of the chromaticity space in the x direction; C′ _b -c _y indicates the projection of the chromaticity space in the y direction.

The constants in the analytical formula are:

C _x = 109.38, _Cy = 152.02, θ = 2.53 (radians), ec _x = 1.60, ec _y = 2.41, a = 25.39, b = 14.03

3. Certification

Although color information is very useful information for face detection, color information can only segment the skin color area of the face, including the face, hands and other backgrounds, reducing the range of face detection, so only color information is used It is not enough to detect the human face, and other follow-up processing is required to confirm whether the human face exists in the skin color area. This embodiment provides a layered processing model for the authentication of the human face candidate area, as shown in Figure 1 , the specific method is as follows:

(1) First, use the above human face skin color model to segment the skin color area, and use median filtering, morphological processing operators, etc. to eliminate non-face areas;

(2) Secondly, in order to deal with faces in different directions, an ellipse is used to approximate the shape of the face, because according to the angle of the major and minor axes of the ellipse, the rotation angle of the face can be analyzed, and the ratio of the major and minor axes of the face candidate ellipse is specified. The size of the major and minor axes is within a certain range, and the ellipse area that is not within this range will be treated as a non-face area.

In this embodiment, an ellipse fitting algorithm based on the least squares method is used to fit the face candidate area: Let F(a, x) be the elliptic curve to be found, where a=[a, b, c, d,e,f] is the ellipse parameter vector, x=[x ² ,xy,y ² ,x,y] is the sample point coordinate polynomial vector; among them, a=[a,b,c,d,e,f] is the ellipse parameter vector, the X-axis component of the x ellipse, and the Y-axis component of the y ellipse.

Define F(a, _xi )=d as the algebraic distance from the sample point _xi to the curve F(a,x)=0. Given N sample points, the solution of the ellipse parameter vector is:

Due to the difference in color between facial features such as eyes, mouth, eyebrows and human face skin in the face area, the texture of the face area is more complex than other face candidate areas such as hand shape and neck, and can be calculated using variance Texture complexity, according to the size of the variance to further remove non-face areas.

Then, the angle θ between the long and short axes of the ellipse is normalized by rotating the face candidate area, and the formula is as follows:

X _rotated = X cos(θ)+Y sin(θ) (13)

Y _rotated = Y cos(θ)-X sin(θ) (14)

X _rotated represents the value after rotating the angle θ in the direction of the long axis of the ellipse; Y _rotated represents the value after rotating the angle θ in the direction of the short axis of the ellipse.

Finally, connected domain analysis: the grayscale and texture of the eyes, mouth, and eyebrows should be clearly distinguished from other areas of the face, which can be clearly seen from the detected face area. Here, firstly, the detected candidate face regions are grayscaled, and after analyzing a large number of face feature images, they are subjected to gradient processing in the Y direction, and then the face template is used for connected domain analysis. When the corresponding When the grayscale sum in the connected domain exceeds a certain threshold, the candidate face area is considered to be a face area. Below a certain threshold, the opposite is true.

Table 2-1 records the detection speed of single face and multiple faces. Here, a video image sequence of 1000 frames is used as the experimental sample. The total time-consuming of single face detection is 21258 milliseconds, and the speed is about 21 milliseconds/frame, while the total time-consuming of multi-face detection is 35666 milliseconds, and the speed is 36 milliseconds/frame. frame, the detection accuracy is 91.8% and 86.6% respectively, because the experiment adopts 25 frame/second video image, the PC of AMD700/128RAM obviously has met the real-time requirement. It can be seen from the experimental data that this algorithm has a good effect.

Test frame number Test time (milliseconds) Detection speed (ms/frame) Detection accuracy single face 1000 21258 21.258 91.8% Multiple faces 1000 35666 35.666 86.6%

The experimental results show that this model can not only detect the frontal face, but also has a good detection effect on facial expressions at any horizontal angle and posture. For various postures, angles, distances, mutual occlusions, etc., it can effectively detect multiple face targets.

Claims (8)

1. a kind of face detection method based on skin color stratification, it is characterized in that, step is: Step 1: Establishment of face skin color model 1.1 Color space selection Use the YCrCb color space to build a human face skin color model; 1.2 Nonlinear transformation of YCrCb color space The chrominance component of the YCrCb color format is not completely independent of the brightness information Y, and the clustering area of the skin color also changes nonlinearly with the brightness information Y; therefore, the nonlinear transformation of the YCrCb color format makes Y The value has as little influence on the chroma component CrCb as possible; 1.3 Ellipse fitting Use an ellipse to approximate the skin color area, and calculate its analytical expression to obtain the face skin color model after ellipse fitting; Step 2: Shape Authentication Establish a hierarchical processing model for the authentication of face candidate areas, the specific method is as follows: 1) Segmentation of skin color regions using the face skin color model; 2) In order to deal with faces in different directions, an ellipse is used to approximate the shape of the face. According to the angle of the major and minor axes of the ellipse, the rotation angle of the face is analyzed, and the ratio of the major and minor axes of the face candidate ellipse is specified. The size of the major and minor axes is set at Within a certain range, the ellipse area that is not within this range will be treated as a non-face area; Step 3: Texture Verification Due to the difference in color between the eyes, mouth, eyebrows, facial features and face skin in the face area, the texture of the face area is more complex than other candidate areas such as hand shape and neck, and the texture complexity is calculated according to the texture complexity To further remove non-face areas; Step 4: Orientation Normalization For the angle between the major and minor axes of the ellipse, rotate the face candidate area to normalize the direction; Step 5: Connected Domain Analysis The grayscale and texture of the eyes, mouth, and eyebrows are obviously different from other areas of the face, which can be clearly seen from the detected face area; Firstly, the detected candidate face area is grayscaled. After analyzing the face feature image, it is processed with a gradient in the Y direction. Y represents the black space, and then the face template is used for connected domain analysis. When When the grayscale sum in the corresponding connected domain exceeds a certain threshold, the candidate face area is considered to be a human face area; when it is lower than a certain threshold, the opposite is true. 2. a kind of face detection method based on skin color layering as claimed in claim 1, is characterized in that: in described step 1, the concrete method of the nonlinear transformation of YCrCb color space is: Perform piecewise linear fitting on the boundary of the chroma component CrCb, and use the boundary of the chroma component CrCb to limit the skin color clustering area. The specific formula is as follows: ${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>\frac{{\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}}{{\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>& \mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; orK</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; Y</span>}\\ {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span>& \mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; \&rsqb;</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$ ${\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; WL</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; WL</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}& \mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}\\ {\mathrm{<span\; class="notranslate">\; WH</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; WH</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}& {\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; Y</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$ ${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 108</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}& \mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}\\ \mathrm{<span\; class="notranslate">\; 108</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}& {\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; Y</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$ ${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 154</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}& \mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}\\ \mathrm{<span\; class="notranslate">\; 154</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}}& {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; Y</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$ C' _i is the transformed chroma component C' _b , C' _r , i is the component of b and r, Denotes the average value of C _i , is the scaling value of the C′ _i component, is the scaling value of the C′ _i component in the Y direction, W is the scaling value, L _Ci is the hue direction of C′ _i , H _Ci is the saturation direction of C′ _i , K ₁ is the offset value of the L component, K _h is the offset value in the H direction, Y _max is the maximum value, and Y _min is the minimum value; After the above-mentioned nonlinear segmented color transformation, and then project it into the Cr'Cb' two-cone space, the human face skin color model is obtained. 3. a kind of human face detection method based on skin color layering as claimed in claim 2, is characterized in that: in described step 1, according to human face skin color sample training estimation value in HHI image storehouse, in YCrCb space: ${\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 46.97</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; WL</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; twenty\; three</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; WH</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 38.76</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; WL</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 20</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; WH</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; ,</span>$ K _l =125, K _h =188, Y _min =16, Y _max =235. 4. a kind of face detection method based on skin color layering as claimed in claim 2, is characterized in that: in described step 1, the formula of ellipse fitting is: $\frac{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; ec</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; +</span>\frac{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; ec</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; b</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$ $\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{cc}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}\end{array}\right]\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}\\ {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$ Among them: ec _x and ec _y are the coordinates of the center of the ellipse; a, b are the major and minor axes; θ is an arbitrary angle of rotation; when an arbitrary angle is rotated, the new elliptic equation becomes formula (10); C′ _b -c _x Indicates the projection of the chromaticity space in the x direction; C′ _b -c _y indicates the projection of the chromaticity space in the y direction. 5. a kind of face detection method based on skin color layering as claimed in claim 4, is characterized in that: in described step 1, in formula (9) and formula (10), C _x = 109.38, _Cy = 152.02, θ = 2.53 radians, ec _x = 1.60, ec _y = 2.41, a = 25.39, b = 14.03. 6. a kind of face detection method based on skin color layering as claimed in claim 1, is characterized in that: in described step 2, when utilizing people's face skin color model to carry out skin color area segmentation, use median filtering, morphology processing The operator eliminates non-face areas. 7. a kind of human face detection method based on skin color layering as claimed in claim 1, is characterized in that: in described step 2, adopt the ellipse fitting algorithm based on least square method, the human face candidate area is fitted . 8. A kind of face detection method based on skin color layering as claimed in claim 1, is characterized in that: in described step 3, use variance to calculate texture complexity, further remove non-human face area according to the size of variance.