CN100345153C - Man face image identifying method based on man face geometric size normalization - Google Patents

Abstract

The invention relates to a face image recognition method based on normalization of face geometric dimensions, and belongs to the technical field of image processing. The method includes: determining the coordinates of the left and right eyeballs on the input face image, and rotating the image to a horizontal position according to the coordinates to obtain a face image 2, and then determining the coordinates of the left and right eyeballs and the mandible point of the face image 2, and specifying the face image The normalized geometric size value of the face image 2 is enlarged or reduced to obtain the face image 3 that meets the standard distance; according to the coordinate positions of the left and right eyeballs and mandible points of the face image 3, the face image 3 is cut A standard normalized face image is obtained. For the training set, known and to-be-recognized face images, form geometrically normalized face images, and extract face features, and use calculation similarity and similarity-based Sort of methods for face recognition. The invention improves the visual effect of the human face, and the recognition rate is greatly improved.

Description

Face Image Recognition Method Based on Normalization of Face Geometric Size

technical field

The invention belongs to the technical field of image processing, in particular to a method for improving the face recognition rate.

Background technique

Face recognition involves many disciplines, including image processing, computer vision, pattern recognition, etc., and is also closely related to the research results of physiology and biology on the structure of the human brain. The recognized difficulties of face recognition are:

(1) Face changes caused by age;

(2) Face diversity caused by posture;

(3) Face plastic deformation caused by expression;

(4) The multiplicity of face patterns caused by factors such as glasses and makeup;

(5) Differences in face images caused by illumination.

In the general face recognition algorithm, there is no normalization process of the standard geometric size of the face image, and the normalization process of the geometric size will not only affect the recognition rate of face recognition, but also affect the number of people in the face database. face visuals. The existing face geometric size normalization method is mainly based on the distance between the two eyes, however, the distance between the two eyes of the face is unstable, especially in the horizontally rotated face image, this Instability is even more prominent.

Contents of the invention

In order to improve the recognition rate of face recognition, the present invention proposes a face image recognition method based on the normalization of the geometric size of the face, which uses the vertical distance from any point on the submandibular line to the line connecting the eyes The size of the image is normalized, and the face features are extracted on this basis, which can improve the recognition rate of the face recognition to a certain extent and improve the visual effect of the face image in the face database.

The present invention proposes a human face image recognition method based on the normalization of the geometric size of the human face, which includes two parts: normalization of the geometric size of the human face and recognition of the human face, and is characterized in that the normalization of the geometric size of the human face Transformation includes the following steps:

1) Determine the coordinate position (x ₁ , y ₁ ) of point A on the left eyeball and the coordinate position (x ₂ , y ₂ ) of point B on the right eyeball on the input face image. Point to make a straight line L ₁ , and determine the coordinates of the mandibular point C ₀ (x ₀ , y ₀ );

2) Calculate the angle α between the straight line L ₁ and the horizontal line;

The included angle α between the straight line L1 and the horizontal line is obtained by formula (1), where (x ₁ , y ₁ ), (x ₂ , y ₂ ) correspond to the coordinates of the left and right eyeballs respectively:

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>$

3) The face image is rotated at a rotation angle of -α to obtain a face image 2;

The rotation expression is as follows:

$\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \text{'}</span>\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \text{'}</span>\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{cc}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}& \mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}& \mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\end{array}\right]\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\end{array}\right]$

In the formula, x, y are the coordinates of the input face image, and x', y' are the coordinates of the face image 2;

4) Determine the coordinate position (x ₃ , y ₃ ) of a point C on the left eyeball and the coordinate position (x ₄ , y ₄ ) of a point D on the right eyeball on the face image 2, and pass through the two points C and D Make a straight line L ₂ , and determine the coordinate position (x ₅ , y ₅ ) of the mandible point E of the face image 2;

5) Specify the numerical value of the geometric size of the normalized face image, where the width is W and the height is H; specify the standard value of the vertical distance from any point on the submandibular line to the line connecting the eyes H ₀ , the standard value of the vertical distance to the lower frame of the image is H ₁ , the standard value of the vertical distance from the line connecting the two eyes to the upper frame of the image is H ₂ ;

6) Calculate the vertical distance h _y from the point E to the straight line L2, and calculate the image scaling factor K=h _y /H ₀ ;

Among them, the vertical distance h _y from point E to straight line L2 is

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}$

7) Enlarge or reduce the face image 2 according to the scaling factor K to obtain a face image 3 satisfying the standard distance _H0 ;

8) Determine the coordinate position (x ₆ , y ₆ ) of a point M on the left eyeball, the coordinate position (x ₇ , y ₇ ) of a point N on the right eyeball, and the longitudinal direction of the mandibular point P on the face image 3. Coordinate y ₈ position;

9) Crop the face image 3 to obtain a standard normalized face image, and cut out the x coordinate in the face image 3 that is less than (x ₆ +x ₇ )/2-W/2 and greater than (x ₆ +x ₇ )/2+W/2, and the part whose y coordinate is less than (y ₇ -H ₂ ) and greater than (y ₈ +H ₁ ). If the width of the cropped image is smaller than W or the height is smaller than H, the interpolation method is used to fill the width to W or the height to H;

Described face recognition comprises the following steps:

10) adopt step 1)～step 9) to form the human face image of geometric dimension normalization to the face image of each person of training set, extract the face feature to the normalized face image;

11) Use steps 1) to 9) to form a face image with normalized geometric dimensions for each known face image, extract face features from the normalized face image, and create a face image containing known faces facial features and compressed images of known faces and databases of personal identification files of known persons;

12) adopt step 1)～step 9) to form the human face image of geometric size normalization for the face image of each person to be identified, and extract the face feature to the normalized face image;

13) In the known face database, the faces to be recognized are recognized by the method of calculating similarity and sorting by similarity.

Steps 10)-13) in the above-mentioned face recognition part can be realized by using existing mature technology.

Features and effects of the present invention

The feature of the present invention is to select the stable key feature points with high precision in the active face as a reference to realize the normalization of the geometric size of the face, achieve the best normalization effect of the face image, and improve the face Visual effects of faces in the database. The recognition rate of the face image recognition method based on the normalization of face geometric size has been greatly improved compared with the common face image recognition algorithm.

Description of drawings

Figure 1 is the input original face image.

FIG. 2 is a schematic diagram of dividing eyeball candidate regions into 9 sub-image regions according to the present invention.

3 is a schematic diagram of the positioning of the two eyes and the mandible and the angle α between the straight line L ₁ and the horizontal line in the original face image.

FIG. 4 is a schematic diagram of the human face image 2 obtained by rotating the original human face image, and the positions of the two eyes and the lower jaw.

FIG. 5 is a schematic diagram of definitions and standards of geometric dimensions of a face image according to an embodiment of the present invention.

FIG. 6 is a schematic diagram of the human face image 3 obtained by scaling the human face image 2 and the positions of the two eyes and the mandible therein.

Fig. 7 is the normalized human face image finally obtained by the present invention.

Fig. 8 is a schematic diagram of extracting five parts of naked face, eyebrows+eyes, eyes, nose tip, and mouth from a normalized face image.

Detailed ways

The embodiment of the face image recognition method based on the normalization of the face geometric size proposed by the present invention is described in detail in conjunction with the accompanying drawings. The method includes the following steps:

1) Determine the coordinate position (x ₁ , y ₁ ) of a point A on the left eyeball and the coordinate position (x ₂ , y ₂ ) of a point B on the right eyeball on the input face image (as shown in Figure 1) ), and make a straight line L ₁ through two points A and B, and determine the coordinates (x ₀ , y ₀ ) of the mandibular point C ₀ .

In this step, the coordinate positions of points A and B on the left and right eyeballs can be realized in two ways. One method is to use the mouse to directly determine the coordinate positions of points A and B on the left and right eyeballs on the face image. It uses an algorithm combining integral projection and feature space analysis to automatically determine the coordinate positions of points A and B on the left and right eyeballs on the face image. This implementation method adopts the second method, which method includes the following steps:

①Detect face area:

Face area detection, that is, to determine the upper, lower, left, and right edge positions of the face area in the image. In this embodiment, the sobel operator is applied to the input image to detect the edge of the image, and the position of the face area is determined by analyzing the integral projection of the edge image in the horizontal and vertical directions. The calculation of the integral projection of the edge map in the horizontal and vertical directions is shown in formulas (1), (2).

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

If point (x,y) is a detected edge point, E(x,y)=1; otherwise E(x,y)=0.

The left and right edges x _l and x _r of the face area are determined by the following formula:

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; max</span>}}\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

In the formula, x ₀ is the x coordinate corresponding to the maximum value of the vertical integral projection of the edge map, that is, the minimum and maximum x values that make the vertical integral projection value greater than 1/3 (empirical value) of the maximum value of the vertical integral projection are used as the face area left and right edges. The upper and lower edges y _t and y _b of the face area are determined by formulas (5), (6):

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ></span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

y _b ＝y _t +(x _r -x _l )×0.8 (6)

Adding 1/3 of the face area width (empirical value) to the formula (5) is to minimize the influence of hair on the positioning results; although this step can only roughly determine the face area in the image, it guarantees Both eyes are included in it.

② Determine eyeball position candidate points:

Candidate points for eyeball locations are determined by analyzing the grayscale and gradient distributions of eye region images. Select the point with the largest difference between the sum of the upper integral projection and the sum of the lower integral projection in the gradient projection histogram as the starting candidate point Y _O of the eyeball ordinate, as shown in formula (7):

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 15</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 15</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

After y _O is determined, this embodiment selects all points within the area of 30 pixels (experience value) above y _O as the initial candidate points of the eyeball position. For each candidate point, examine the gray distribution of the 30×30 image area centered on this point. In this embodiment, the image area is divided into 9 sub-image areas as shown in FIG. 2, and the gray-scale integral of each sub-image area is calculated, as shown in formula (8).

Here I(x, y) represents the gray value at the point (x, y). Since the gray value of the eyeball is generally smaller than the gray value of its surrounding area, if the gray integral si of a certain sub-image area, _i =1, 2, 3, 4, 6, 7, 8, 9 is less than The central sub-image, that is, the gray integral S ₅ of region 5, removes this candidate point. The remaining candidate points are used as candidate points for the final eyeball position.

Due to the reflection of the eyeball, many face images will leave small bright spots inside the eyeball, which will cause some good candidate points to be removed by mistake. Therefore, before calculating the gray integral projection, it is necessary to remove small bright spots on the face image. For each point in the face image, replace its gray value with the minimum gray value of 9 points within a 3×3 neighborhood centered on it.

③ Determination of eyeball position:

For the detected candidate points, the present invention uses the feature space analysis method (PCA) to determine the final eyeball position. The eye region images in another 9 groups of face images with different poses are selected as the training set (the eye region can be determined by manual positioning), and 18 feature spaces of the left eye and right eye are trained respectively. For each candidate point C _i of the eyeball position, its corresponding sub-image is projected into these 18 groups of feature spaces to obtain the projection vector P _i , i=1, 2...18, and the matching error of each projection vector is defined as:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; D.</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

Here p _k is the value of the k-th dimension of the projection vector, λ _k is the eigenvalue corresponding to the k-th eigenvector, and D is the dimension of the retained eigenvector. The matching error of each feature point corresponding to the image area is defined as the minimum value of the matching error of its projection vector in 18 groups of feature spaces. For all candidate points, the candidate point with the smallest matching error is selected as the position of the first eyeball. In order to prevent both eyeballs from being located on the same eye, select the point with the smallest matching error among the candidate points whose distance from the first eyeball position is greater than a given value as another eyeball position. After these two positions are determined, select the left eyeball A (x ₁ , y ₁ ) with the smaller x coordinate, and choose the right eyeball B (x ₂ , y ₂ ) with the larger x coordinate

④ Determine the coordinates (x ₀ , y ₀ ) of the mandibular point C ₀ : two methods can be used. The first method uses the mouse to directly determine it on the face image, and the second method uses integral projection and the proportion of facial organs (parts) combined method.

This embodiment adopts the method of integrating integral projection and facial organ (part) ratio, that is, the mandibular point is determined according to the horizontal integral projection map of the human face area, the position of the eyeball and the proportional relationship of various organs on the human face. The method includes two steps: one is to detect candidate valley points that may correspond to organs (that is, the valley points of the horizontal integral projection curve), and the other is to determine which candidate valley points correspond to each organ.

Mean filtering is performed on the transverse integral projection curve first, and then its second order derivative is calculated. The extreme points of the second derivative correspond to the peak and valley points of the projection curve, and the maximum points correspond to the valley points of the projection curve. Detect the peaks of the second derivatives as the candidate locations of the ordinates of the facial organs. For the obtained candidate positions, a loop strategy is used to find out the situation that best matches the proportion of human face organs as the final result. In this way, the ordinates of each organ are obtained, including the ordinate _{y 0 of} the mandibular point C 0 . The abscissa x ₀ of C ₀ is determined by the midpoint of A and B, that is, x ₀ =(x ₁ +x ₂ )/2. As shown in Figure 3, the three points A, B, and _C0 are the determined coordinate points of the left and right eyes and the mandibular point respectively. In the figure, the straight line _L1 is the connecting line between A and B, _L0 is the horizontal line, and the straight line L ₁ and the horizontal line L ₀ the angle α.

2) Calculate the angle α between the straight line L ₁ and the horizontal line L ₀ .

The angle α between the straight line L1 and the horizontal line L ₀ is obtained by the formula (10), where A(x ₁ , y ₁ ), B(x ₂ , y ₂ ) correspond to the coordinates of the left and right eyeballs respectively:

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

3) Carry out the rotation processing that the rotation angle is -α to this face image, obtain face image 2, as shown in Fig. 4, in Fig. 4, C, D are respectively left eyeball point and right eyeball point, straight line L ₂ is the connecting line passing through C and D, E is the mandibular point, _L3 is the horizontal submandibular line, h _y is the vertical distance from E point to the straight line L2.

Let the width and height of the input face image be SrcWidth and SrcHeight in turn, and the width and height of the face image 2 obtained through rotation are w and h respectively.

Since there is a translation of the coordinate origin in the actual image rotation, it is necessary to perform offset correction on the coordinates. Let the offsets in the horizontal and vertical directions be dx and dy respectively. According to the difference of α, there is the following relationship:

When α>0:

     w=INT(SrcWidth×cosα+SrcHeight×sinα);

h＝INT(SrcWidth×sinα+SrcHeight×cosα);

dx=0;

dy = SrcWidth×sinα; (10.1)

When α<0:

     w=INT(SrcWidth×cosα-SrcHeight×sinα);

h＝INT(-SrcWidth×sinα+SrcHeight×cosα);

dx＝-SrcHeight×sinα;

dy=0; (10.2)

For each point (i, j) in the face image 2, let the corresponding point (i, j) in the original image be (io, jo). when $\left\{\begin{array}{c}0\&le;\mathrm{io}<\mathrm{SrcWidth}\\ 0\&le;\mathrm{jo}<\mathrm{SrcHeight}\end{array}\right.$ , it indicates that there is a point (io, jo) corresponding to (i, j) in the original image, considering the offset dx, dy, there are:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; io</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; dx</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; dy</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; jo</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; dx</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; dy</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

Then Image2[j][i]=OriginalImage[jo][io]; otherwise, (io, jo) is located in the blank part of Image 2, which can be assigned as: Image2[j][i]=0.

4) Determine the coordinate position (x ₃ , y ₃ ) of a point C on the left eyeball and the coordinate position (x ₄ , y ₄ ) of a point D on the right eyeball on the face image 2, and pass through the two points C and D Make a straight line L ₂ , and determine the coordinate position (x ₅ , y ₅ ) of the mandible point E of the face image 2;

In this step, the coordinate positions of points C and D on the left and right eyeballs can be realized by three methods. The first method is to use the mouse to directly determine the coordinate positions of points C and D on the left and right eyeballs on the face image. The first method is to use the same method for determining the coordinates of A and B to determine the coordinate positions of points C and D on the left and right eyeballs on the face image. The third method is to calculate the coordinates of C and D through the coordinates of A and B and α. The coordinates of the point. This embodiment adopts the third method:

x ₃ =INT(x ₁ ×cosα+y ₁ ×sinα+dx+0.5);

y ₃ =INT(-x ₁ ×sinα+y ₁ ×cosα+dy+0.5);

_x4 ＝INT( _x2 ×cosα+ _y2 ×sinα+dx+0.5);

y ₄ =INT(-x ₂ ×sinα+y ₂ ×cosα+dy+0.5); (12)

In this step, the coordinate position (x ₅ , y ₅ ) of the mandible point E of the face image 2 can be realized by three methods. The first method is to use the mouse to directly determine the coordinate position of the mandible point E on the face image , the second method is to use the same method of determining the mandibular point C ₀ to determine the coordinate position of the upper and lower jaw point E of the face image, and the third method is to calculate the coordinates of the E point through the coordinates of C ₀ and α (x ₅ , y ₅ ). This embodiment adopts the third method:

_x5 ＝INT( _x0 ×cosα+ _y0 ×sinα+dx+0.5);

y ₅ =INT(-x ₀ ×sinα+y ₀ ×cosα+dy+0.5); (13)

5) Specify the numerical value of the geometric size of the face image normalized by the geometric size, as shown in Figure 5, in Figure 5, 1 is the upper frame of the image, 2 is the straight line _L2 determined by the two eyes, and 3 is the submandibular line L ₃ , 4 is the lower border of the image, 5 is the right border of the image, and 6 is the left border of the image.

In this embodiment, it is specified that the width of the image (the distance between 5 and 6) is W=360 pixels, and the height (the distance between 1 and 4) is H=480 pixels. The standard value of the vertical distance between any point on the submandibular line and the line between the two eyes (the distance between 2 and 3) is H ₀ =200 pixels, and the vertical distance to the lower frame of the image (the distance between 3 and 4 ) has a standard value of H ₁ =28 pixels, and the standard value of the vertical distance from the line connecting the two eyes to the upper frame of the image (the distance between 1 and 2) is H ₂ =252 pixels.

6) Calculate the vertical distance h _y from point E to straight line L2 as

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$

And calculate the image scaling factor K= _hy /H ₀ .

7) The face image 2 is enlarged or reduced according to the scaling factor K, and the face image 3 meeting the standard distance _H0 is obtained, as shown in Figure 6, and Figure 6 is the figure after the reduction process of Figure 4, in the figure , M and N are the coordinate points of the left eyeball point and the right eyeball point, _L4 is the submandibular horizontal line, and P is the mandibular point.

The width (w ₃ ) and height (h ₃ ) of the face image 3 are:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$

8) Determine the coordinate position (x ₆ , y ₆ ) of a point M on the left eyeball, the coordinate position (x ₇ , y ₇ ) of a point N on the right eyeball, and the longitudinal direction of the mandibular point P on the face image 3. coordinate location.

Then the coordinates of the midpoint MidPoint of the two eyes M and N satisfy the following formula:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 6</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 7</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 6</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 7</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; INT</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span>$

The ordinate y ₈ of point P satisfies:

y ₈ =MidPoint.y+H ₀ (17)

9) Cutting the face image 3 to obtain a standard normalized face image.

According to the size of the standard normalized image, cut off the part of the face image 3 whose x coordinate is less than (x ₆ +x ₇ )/2-W/2 and greater than (x ₆ +x ₇ )/2+W/2, And the part whose y coordinate is less than (y ₇ -H ₂ ) and greater than (y ₈ +H ₁ ).

In specific implementation, the cropping rectangle CropRect is defined, and the coordinates of its left and right boundaries and upper and lower boundaries are respectively defined as CropRect.left, CropRect.right, CropRect.top, and CropRect.bottom. The specific cut size is:

$\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; CropRect</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; left</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ?</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; :</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; CropRect</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; right</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; ?</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\frac{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; :</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; CropRect</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; top</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ?</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; MidPoint</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; :</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; CropRect</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; bottom</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; <figure-callout\; id="8"\; label="y"\; filenames="C20051006796200142.png"\; state="\{\{state\}\}">y</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 8</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; ?</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; <figure-callout\; id="8"\; label="y"\; filenames="C20051006796200142.png"\; state="\{\{state\}\}">y</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 8</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; :</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 18</span>}<span\; class="notranslate">\; )</span>$

The face image 3 is cropped according to the cropping size. If the width of the cropped image is smaller than W or the height is smaller than H, the interpolation method is used to fill the width to W or the height to H to obtain a normalized face image of a standard size, as shown in Figure 7.

10) For each person's face image in the training set, adopt steps 1) to 9) to form a geometrically normalized face image, and extract each person's naked face image for each normalized face image. Face, eyebrows+eyes, eyes, nose tip, and mouth five face parts, for the five face parts extracted from the training set faces, the naked face, eyebrows+eyes, eyes, nose tip, and mouth, use the principal component The eigenface method in the analysis method respectively forms the characteristic naked face, characteristic (eyes + eyebrows), characteristic eyes, characteristic nose, and characteristic mouth. Figure 8 shows examples of these five components. The feature extraction and recognition algorithm of this embodiment adopts the patent No. 01136577.3: multi-mode face recognition method based on component principal component analysis.

11) For each known human face image, adopt steps 1) to 9) to form a geometrically normalized human face image, and extract the naked face and eyebrows from the known normalized human face +Five face parts including eyes, eyes, nose tip, and mouth, use the characteristic projection value analysis method in the principal component analysis method to extract five types of human faces including naked face, eyes+eyebrows, eyes, nose, and mouth. The projected feature values of the parts and the establishment of a database including the projected feature values of the five face parts of known faces, naked face, eyes+eyebrows, eyes, nose, and mouth, compressed images of known faces, and personal identity files of known people. The feature extraction and recognition algorithm in this step also adopts the patent No. 01136577.3: multi-mode face recognition method based on component principal component analysis.

12) Use steps 1) to 9) to form a normalized face image of geometric size for the face image of each person to be recognized, and extract the naked face, eyebrows+eyes from the normalized face to be recognized Using the projection eigenvalue analysis method in the principal component analysis method to extract the projection eigenvalues of the naked face, eyes+eyebrows, eyes, nose, and mouth of the face to be recognized. The feature extraction and recognition algorithm in this step also adopts the patent No. 01136577.3: multi-mode face recognition method based on component principal component analysis.

13) The method of global face recognition and partial face recognition is used for face recognition. The process of face recognition is: compare the features of the face to be recognized with the features of the face stored in the database, calculate the similarity, and then compare the faces in the database according to the similarity with the face to be recognized. To the small order, and in this order, display the photo of the person being found, the personal identity file and the similarity with the person to be identified, so as to find out the identity of the person to be identified or similar in appearance to the person to be identified person's identity. Formula (19) is used to calculate the similarity between the face to be recognized and the known face.

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; =</span>\sqrt{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 19</span>}<span\; class="notranslate">\; )</span>$

(19) where A is the feature projection value string of the face to be recognized, and B is the normalized feature projection value string of the known face in the database.

When this embodiment adopts the global face recognition method, the feature projection values of the naked face, eyes+eyebrows, eyes, nose, and mouth of the known face are weighted in a ratio of 5:6:4:3:2, and at the same time The feature projection values of bare face, eyes+eyebrows, eyes, nose, and mouth are weighted according to the ratio of 5:6:4:3:2, and then the similarity is calculated according to (19).

When using the partial face recognition method, use the method of human-computer interaction to select any combination of naked face, eyes+eyebrows, eyes, nose, and mouth, and the number of combinations is 5! = 120 types, that is, there are 120 face recognition modes in total. The feature projection values of bare face, eyes+brows, eyes, nose, and mouth are still weighted in a ratio of 5:6:4:3:2. This step also adopts the patent No. 01136577.3: multi-mode face recognition method based on component principal component analysis.

Claims (10)

1. A human face image recognition method based on normalization of human face geometric size, comprising two parts of normalization of human face geometric size and face recognition, characterized in that the normalization of human face geometric size Include the following steps: 1) Determine the coordinate position (x ₁ , y ₁ ) of point A on the left eyeball and the coordinate position (x ₂ , y ₂ ) of point B on the right eyeball on the input face image. Point to make a straight line L ₁ , and determine the coordinates of the mandibular point C ₀ (x ₀ , y ₀ ); 2) Calculate the angle α between the straight line L ₁ and the horizontal line; The angle α between the straight line L1 and the horizontal line is given by $\mathrm{\&alpha;}=\arctan \left(\frac{{\mathrm{the\; y}}_2-{\mathrm{the\; y}}_1}{x_2-x_1}\right)$ obtained by the formula, Where (x ₁ , y ₁ ), (x ₂ , y ₂ ) correspond to the coordinates of A and B points of the left and right eyeballs respectively: 3) Carrying out the rotation processing that the rotation angle is -α to this human face image, obtains the second human face image; The rotation expression is as follows: $\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\\ {\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{cc}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}& \mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\\ <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}& \mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&alpha;</span>}\end{array}\right]\left[\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\end{array}\right]$ In the formula, x, y are the coordinates of the point on the input face image, and x', y' are the coordinates of the point on the second face image; 4) Determine the coordinate position (x ₃ , y ₃ ) of a point C on the left eyeball and the coordinate position (x ₄ , y ₄ ) of a point D on the right eyeball on the second face image. Point to make a straight line L ₂ , and determine the coordinate position (x ₅ , y ₅ ) of the mandible point E of the second face image; 5) Specify the numerical value of the geometric size of the normalized face image, where the width is W and the height is H; specify the standard value of the vertical distance from any point on the submandibular line to the line connecting the eyes H ₀ , the standard value of the vertical distance to the lower frame of the image is H ₁ , the standard value of the vertical distance from the line connecting the two eyes to the upper frame of the image is H ₂ ; 6) Calculate the vertical distance h _y from the point E to the straight line L ₂ , and calculate the image scaling factor K=h _y /H ₀ ; Among them, the vertical distance h _y from point E to straight line _L2 is ${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}{<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}$ 7) Enlarge or reduce the second human face image according to the scaling factor K to obtain the third human face image satisfying the standard distance _H0 ; 8) Determine the coordinate position (x ₆ , y ₆ ) of a point M on the left eyeball, the coordinate position (x ₇ , y ₇ ) of a point N on the right eyeball, and the point P of the mandible on the third face image Ordinate y ₈ position; y ₈ =MidPoint.y+H ₀ ; 9) Cutting the third face image to obtain a standard normalized face image, cutting out the x coordinates in the third face image less than (x ₆ +x ₇ )/2-W/2, greater than (x ₆ + x ₇ )/2+W/2, and the part whose y coordinate is less than (y ₇ -H ₂ ) and greater than (y ₈ +H ₁ ); if the width of the cropped image is less than W or the height is less than H, use Interpolation method, fill width to W or height to H; Described face recognition comprises the following steps: 10) adopt step 1)～step 9) to form the human face image of geometric dimension normalization to the face image of each person of training set, extract the face feature to the normalized face image; 11) Use steps 1) to 9) to form a face image with normalized geometric dimensions for each known face image, extract face features from the normalized face image, and create a face image containing known faces facial features and compressed images of known faces and databases of personal identification files of known persons; 12) adopt step 1)～step 9) to form the human face image of geometric size normalization for the face image of each person to be identified, and extract the face feature to the normalized face image; 13) In the known face database, the faces to be recognized are recognized by the method of calculating similarity and sorting by similarity. 2. The human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, characterized in that, determining the coordinate positions of points A and B on the left and right eyeballs in the step 1) is to use the mouse to directly Read out the coordinate positions of points A and B on the left and right eyeballs on the face image. 3. The human face image recognition method based on the normalization of human face geometric dimensions as claimed in claim 1, characterized in that, determining the coordinate positions of points A and B on the left and right eyeballs in the step 1) adopts integral projection The method combined with feature space analysis determines the coordinate positions of points A and B on the left and right eyeballs on the face image. 4, the human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, is characterized in that, the coordinate position of point C, D on the left and right eyeballs in the described step 4) adopts mouse to directly click on the human face. The coordinate positions of points C and D on the left and right eyeballs are read out from the face image. 5. The human face image recognition method based on the normalization of human face geometric dimensions as claimed in claim 1, wherein the coordinate positions of points C and D on the left and right eyeballs in the step 4) adopt integral projection and feature The method of combining spatial analysis determines the coordinate positions of points C and D on the left and right eyeballs on the face image. 6. The human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, characterized in that, in the step 4), the coordinate positions of points C and D on the left and right eyeballs are adopted by A, B The coordinates and α are calculated to calculate the coordinates of C and D points. 7, the human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, is characterized in that, in described step 1) and step 4), determine mandibular point C ₀ and E coordinate position to adopt mouse to directly Read out on face images. 8, the human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, is characterized in that, in described step 1) and step 4), determine mandibular point C ₀ and E coordinate position and adopt integral projection The method combined with the proportion of human face organs includes two steps: one is to detect the candidate valley points corresponding to the organs, and the other is to determine which candidate valley points correspond to each organ. 9, the human face image recognition method based on the normalization of human face geometric size as claimed in claim 1, is characterized in that, in described step 4), determine the coordinate position of mandibular point E by the coordinate and α of C ₀ , calculate The coordinates of point E. 10. The face image recognition method based on the normalization of geometric dimensions of the face as claimed in claim 1, characterized in that, in said step 5), the numerical value of the geometric dimensions of the human face image normalized by geometric dimensions is specified, Wherein the dimension of width W=360 pixels, the dimension of height is H=480 pixels; the standard value of the vertical distance from any point on the submandibular line to the line connecting the two eyes is H ₀ =200 pixels, and the vertical distance to the lower frame of the image The standard value of the distance is H ₁ =28 pixels, and the standard value of the vertical distance from the line connecting the two eyes to the lower frame of the image is H ₂ =252 pixels.

Images