CN101339669A - Three-dimensional human face modelling approach based on front side image - Google Patents

Abstract

The invention relates to the technical field of computer graphics and image processing, in particular to a 3D human face modeling method based on front silhouette, which mainly models the 3D human face in a fully automatic manner. The method comprises the following steps: a general human face grid model is built by 113 feature points and 184 tri patch information arrays which are connected with the 113 feature points based on Candide parameterization; 2D position information and depth information of a specific human face are determined by virtue of human eyes location algorithm and side nose tip location algorithm by inputting two 2D images, the general face model is modified and transformed into a specific human face model; an interpolating is calculated for the human body patches by Delaunay Triangulation Technology and texture mapping is carried out by the Bilinear Interpolation Technology, thus causing the human face image to be appeared more delicate and accurate; and the method solves the problems such as low efficiency, high calculation expense, and the like, existing in the prior art.

Description

Three-dimensional face modeling method based on positive silhouette

Technical field

The present invention relates to the computer graphic image processing technology field, specifically refer to a kind of method of the three-dimensional face modeling based on positive silhouette.

Background technology

Along with the continuous development of computer graphics, three-dimensional manufacturing technology is also developed rapidly and is widely applied to social production, various fields in life.Nowadays computer network high speed development and come into our daily life is especially for the 3-D technology development with use unlimited living space is provided.Along with the increasingly extensive of network application popularized, from talking Face to face, making a phone call, to Internet chat, send e-mails mutually, people more and more trend towards exchanging in virtual network world.People's face is the human important channel that exchange mutually, is the carrier of complex expression such as human pleasure, anger, sorrow, happiness and language.Along with computer graphics in modeling, play up and the development of aspect such as real-time animation, modeling of people's face and animation have been obtained at aspects such as film virtual role, teleconference, criminology, medical science, information help, man-machine interaction, amusement, virtual reality, recognition of face and expression understandings widely and have been used.The excellent properties performance of computing machine aspect the processing graphics image obtained the people's of a lot of imaginative and creativities attention, thereby promoted the drafting of real more virtual environment.In order to reflect the activity of user in virtual world, improve user's feeling of immersion therein,---complete graphic entity that looks like true man by computer representation---is true to nature more good more just to require the visual human, and its matter of utmost importance is exactly the establishment of moulding of people's face and expression animation.In view of the above, a variety of dimensional Modeling Technology are arranged at present, but great majority all are based on interactively method, need to carry out choosing of key point by the professional, therefore, work efficiency is lower, simultaneously, need choose up to a hundred key points mostly, thereby reach visually satisfied result based on interactively dimensional Modeling Technology, so also there is the big problem of computing cost in this method.

Summary of the invention

The present invention is the above-mentioned disappearance that exists based on interactively method of solution, and the three-dimensional face modeling method of the full-automatic mode of employing that proposes.A kind of three-dimensional face modeling method based on positive silhouette, it comprises the three-dimensional specific people's face modeling based on quadrature image (Orthogonal Images) and three-dimensional general faceform (Generic Face Model).On the initial basis that the faceform is more coarse, dough sheet is less, utilize the triangulation technology, realize exquisitenessization and polishing, and raise the efficiency compromise with computing cost specific faceform.At last, synthesized three-dimensional specific faceform more true to nature.It comprises: the Candide-3 model that the present invention uses is formed the general face wire frame model of structure by 113 summits and 184 tri patchs, and build input by two bidimensional images, determine the two-dimensional position information and the depth information of specific people's face by human eye location algorithm and side nose location algorithm, thereby general faceform is revised, be transformed into specific faceform; By people's face sheet being carried out interpolation calculation, make people's face look fine and smooth more and accurate thereby reach based on shape analysis.

Description of drawings

Fig. 1 is a Candide3 faceform XY structural representation of the present invention;

Fig. 2 is a Candide3 faceform YZ structural representation of the present invention;

Fig. 3 is based on the triangular facet subdivision synoptic diagram of shape analysis;

Fig. 4 cylindrical surface projecting schematic diagram.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described

1 sets up the parameterized common people's face of Candide grid

The present invention adopts Ahlberg to be distributed on up-to-date version Candide-3 faceform in calendar year 2001.This version faceform has reached the unification with the MPEG-4 standard, model comprises 113 key points and 168 tri patchs, controlled variable by static parameter (FacialDefinition Parameters, FDP) and dynamic parameter (Facial AnimationParameters FAP) forms.The present invention uses the Candide-3 model to be made up of 113 summits and 184 tri patchs.

2 general faceforms are to specific faceform's conversion

2.1 image pre-service

In the reality, two width of cloth images (photo) of acquisition can not be big or small the same, so before feature point for calibration, at first will align silhouette (photo) and carry out standardization processing, the head size of two width of cloth images (photo) equated.

2.2 specific face characteristic dot generation

1) border, people's face left and right sides determines

If handled image be I (x, y), its size is M * N, then the vertical Gray Projection function of this image is:

$\mathrm{PV}\left(x\right)=\underset{y=1}{\overset{N}{\mathrm{\Σ}}}I(x,y)$

In the formula, PV is called vertical Gray Projection curve.Observe the vertical Gray Projection curve of different single humanoid figure pictures, can will make vertical Gray Projection curve form a protruding peak in finder's face region with certain width.The border, the left and right sides at this protruding peak roughly is exactly the border, the left and right sides of people's face, and this is that human face region often has higher brightness because compare with background.At people's face left and right sides boundary, the summation of brightness value reduces rapidly on the vertical direction, thereby forms a tangible protruding peak.Therefore, only need to determine the border, the left and right sides at main protruding peak in the vertical Gray Projection curve, can obtain the border, the left and right sides of people's face.In order to remove The noise, vertical Gray Projection curve is carried out smoothing processing with median filter method, the smoothing processing function is:

$\mathrm{PVF}\left(x\right)=\frac{1}{K+1}\underset{i=x-K/2}{\overset{x+K/2}{\mathrm{\Σ}}}\mathrm{PV}\left(i\right)$

In the formula, K is the width of filter window, and its value is relevant with the size of people's face in image, and here, the K value is 6.

2) the human eye horizontal level estimates

After the border, the left and right sides of people's face was determined, the size of establishing border, the left and right sides was m, and then Ci Shi image size is m * N, with quadrat method to the horizontal Gray Projection function of this image-region was:

$\mathrm{PH}\left(x\right)=\underset{x=1}{\overset{m}{\mathrm{\Σ}}}I(x,y)$

$\mathrm{PHF}\left(x\right)=\frac{1}{L+1}\underset{i=y-L/2}{\overset{y+L/2}{\mathrm{\Σ}}}\mathrm{PH}\left(i\right)$

The value of L is relevant with the size of people's face in image, gets L=4 in the experiment.

3) estimation in human eye segmentation threshold interval and accurately location

Obtain the looks area I ₀After, only surplus eyebrow and the eyes that need be distinguished can therefrom separate human eye by seeking optimal segmenting threshold.So-called optimal segmenting threshold refers to can be with iris, pupil, go up the gray scale segmentation threshold that eye frame and eyebrow obviously separate.Because the gray-scale value at iris, pupil and last eye frame position is obviously low than its adjacent domain (circumference of eyes skin) gray-scale value, therefore can cut apart iris, pupil and go up that the gray-scale value at eye frame position always exists, and be not unique, exist certain continuously among a small circle.

Estimate initial threshold T ₀, need detect first peak dot of smoothed histogram.Because the histogram after level and smooth two peak dots seldom occur in continuous 10 gray levels, so its preceding 250 gray levels all can be divided into 25 intervals, obtain the minimum and maximum value of 10 gray levels in each interval, search for the position of first peak dot by low gray area to high gray area then, estimate initial threshold T thus ₀

4) adjust threshold value automatically and determine eye position

T ₀An initial threshold that human eye might be split from image is provided, but optimal segmenting threshold not necessarily.In order to find optimal segmenting threshold, can select a suitable step-length T _Step, determine optimal threshold by increasing step-length.At first, use T ₀To image I ₀Binaryzation judges that whether the human eye black patch occurs, if black patch do not occur, adds a step-length T on the basis of last threshold value _Step, occur up to the eyes black patch, and black block in the binary image is carried out mark, calculate every area (pixel count), determine every occupied rectangle position and wide height.Utilize above-mentioned segmentation threshold to increase progressively, detect and locate the whole process of eyes, wherein try to achieve initial threshold T ₀=60, get T in the experiment _Step=15, when threshold value T=105, the eyes black patch is detected, thereby reaches the accurate location of human eye.

3 Delaunay triangulations based on shape analysis

Specific people's face be one by 113 unique points and 184 three-dimensional face grids that tri patch is formed.Specifically, this people's face grid is also far from the target of three-dimensional face model true to nature.In order to make synthetic specific people's face more and more true to nature, method of the present invention is by the tri patch in specific faceform's rough region is carried out subdivision, insert non-unique point then in current specific faceform's vertex set, carry out interpolation iterative process next time again, circulation has enough visual vivid degree up to this zone successively.If Pi_tris set stored all dough sheets in the dissatisfied zone of visual effect, the step of its triangulation is as follows:

1) shape of calculating tri patch, adopt diverse ways to carry out the subdivision operation then:

At first calculate longest edge E as front piece Ti ₁, vice-minister limit E ₂, minor face E ₃Length is respectively e _I1, e _I2And e _I3Calculate:

$r_1=\frac{e_{i1}}{e_{i2}},$ $r_2=\frac{e_{i2}}{e_{i3}},$ $r_3=\frac{e_{i1}}{e_{i3}}$

2) for three kinds of different shapes, take different processing modes:

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃In the time of≤H, H gets 1.3 in the experiment, at this moment the subdivision algorithm such as the accompanying drawing 3 (a) of tri patch.Wherein, E, F and G point are respectively limit E ₁, E ₂And E ₂Mid point.With E, F and G join among the non-unique point set NNF that subdivision produces as non-unique point, use as the non-unique point in the interpolation iterative process next time, and mark is with AC in markedt simultaneously, and BC and AB are the limit in other tri patchs on limit.The ABC that deletes patch in the dough sheet set adds dough sheet AEF, BEG, and FCG then.

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃When≤H can not satisfy, for the sake of simplicity, we directly found longest edge E ₁Mid point E, judge then and work as r ₂〉=H ₁The time, then triangulation such as Fig. 3 (b).Experimental selection H ₁=1.6 effects are best.This moment, the F point was limit E ₂Mid point.With B, C, E, F spot projection utilize Delaunay trigonometric ratio algorithm then on XOY plane, return two tri patch BEC and ECF behind the trigonometric ratio.Same non-unique point E and the F of adding adds the limit among dough sheet and the mark markedt.

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃≤ H can not satisfy, and r ₂＜H ₁The time, the triangulation of this moment is shown in accompanying drawing 3 (c).Illustrate that tri patch is longer and narrower this moment, and we find longest edge E ₁Mid point E, then the E point is added among the NNF as non-unique point.Be the limit of adding among dough sheet and the mark markedt then.

3) concrete subdivision process is as follows: select each the tri patch Ti among the Pi_tris successively, at first in the first round processing, temporary transient two kinds of tri patchs not considering shown in accompanying drawing 3a and accompanying drawing 3b, the subdivision first round is handled all tri patchs with shape shown in the accompanying drawing 3c.Judge the mark situation on Ti each limit in markedt,, then use the method subdivision shown in accompanying drawing 3a certainly if minor face or vice-minister limit have been labeled.If do not have the limit to be labeled or longest edge is labeled, then use method subdivision shown in accompanying drawing 3c.Second takes turns in the processing, considers the tri patch of two kinds of shapes shown in accompanying drawing 3a and accompanying drawing 3b.Equally, if minor face is labeled, then use the method subdivision shown in accompanying drawing 3a certainly.If minor face is not labeled, then look a certain method subdivision of the concrete shape of tri patch in selecting shown in accompanying drawing 3a or accompanying drawing 4b.

4 texture

The step of carrying out texture mapping may be summarized to be:

Definition texture image, control texture, explanation texture mapping mode, definition texture coordinate.The definition texture image comprises form, the texture image data type of the length of resolution, texture image of given texture image and width, texture image, the memory address of texture image.

The present invention adopts cylindrical surface projecting to realize mapping between the position of the coordinate of three-dimensional face model and 2 d texture picture.

The cylindrical surface projecting principle is drawn a ray as shown in Figure 4 through 1 P of body surface from the center of cylinder, this ray and periphery intersect at a P ' (u, v), some P ' puts the projection of P on cylinder.The cylinder radius r is known.U in the cylindrical coordinates, v represent coordinate components and the ray and the formed in the counterclockwise direction deflection angle of Z axle negative sense of P ' in the vertical direction respectively.If cylindrical center's coordinate is (x ₀, y, z ₀), can get the cylindrical surface projecting formula by accompanying drawing 4 and be:

$\left\{\begin{array}{c}u=y\\ v=\left\{\begin{array}{c}\arctan \frac{x_0-x}{z_0-z},(x\&le;x_0,z<z_0)\\ \frac{\mathrm{\&pi;}}{2},(x\&le;x_0,z=z_0)\\ \mathrm{\&pi;}+\arctan \frac{x_0-x}{z_0-z},(z>z_0)\\ \frac{3\mathrm{\&pi;}}{2},(x>x_0,z=z_0)\\ 2\mathrm{\&pi;}+\arctan \frac{x_0-x}{z_0-z},(x>x_0,z<z_0)\end{array}\right.\end{array}\right.$

For ease of subsequent treatment, the face of cylinder is launched into the plane, can be further cylindrical coordinates (u v) is converted to size and is the coordinate in the rectangle plane of W * H (x ', y '), and conversion formula is:

$\left\{\begin{array}{c}{x}^{\&prime;}=W\&times;\frac{v}{2\mathrm{\&pi;}}\\ y^{\&prime;}=H\&times;\frac{u-y_{\min }}{y_{\max }-y_{\min }}\end{array}\right.$

The border is: u=y _Min, y '=0; U=y _Max, y '=H; V=0, x '=0; V=2 π, x '=W.In order to extract the information on positive and the silhouette (photo) comprehensively, the texture value that defines certain some P is this weighted mean corresponding to the texture value (colouring information) on the positive side projection.If the texture value of P corresponding point in front projection is I _f, the texture value of corresponding point is I on side projection _p, then the texture value I of this point is:

I＝kI _f+(1-k)I _p

K is a weight factor, and when three-dimensional point only had corresponding point in front projection, k got 1; When three-dimensional point only had corresponding point on side projection, k got 0; When three-dimensional point had both had corresponding point in front projection, when on side projection, corresponding point being arranged again, need get a suitable k value.Because front projection is perpendicular to the Z axle, the area of establishing a bin on the three-dimensional model is S, and the normal vector of this bin is n (n _x, n _y, n _z), this bin area S in front projection then _fCan try to achieve by following formula:

S _f＝|n _z|S

Can get S with method _p=| n _x| S.So k can be taken as:

$k=\frac{S_f}{S_f+S_p}=\frac{{|n}_z|}{\left|n_z\right|+\left|n_x\right|}$

As previously mentioned, the value that can get the texture weight factor k of every some P is:

In sum, the present invention has realized the three-dimensional specific people's face modeling based on quadrature image (Orthogonal Images) and three-dimensional general faceform (Generic Face Model).On the initial basis that the faceform is more coarse, dough sheet is less, utilize the triangulation technology, realize specific faceform's exquisitenessization and polishing, and implementation efficiency and computing cost is compromise.At last, synthesized three-dimensional specific faceform more true to nature.

Claims (6)

1. three-dimensional face modeling method based on positive silhouette, comprise: by the parameterized general face wire frame model of Candide is to make up basic people's face grid by 113 unique points with 184 tri patchs that are connected these points, specific face characteristic dot generation, general faceform is to specific faceform's conversion, the specifiable lattice model is carried out interpolation generate meticulousr grid model, use the pixel of X-Y scheme that grid is carried out pinup picture and carries out the texture step, it is characterized in that the generation of described specific human face characteristic point mainly is to adopt based on the eye locating method of Gray Projection function and nose location algorithm eyes and the nose to bidimensional image to position.

2. the three-dimensional face modeling method based on positive silhouette as claimed in claim 1 is characterized in that, described employing positions based on the eye locating method of Gray Projection function and nose location algorithm eyes and the nose to bidimensional image, comprising:

(1) image pre-service;

(2) specific face characteristic dot generation:

1) border, people's face left and right sides determines

If handled image is I:x, y, its size is M * N, then the vertical Gray Projection function of this image is:

$\mathrm{PV}\left(x\right)=\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}I(x,y)$

In the formula, PV is called vertical Gray Projection curve, observe the vertical Gray Projection curve of different single humanoid figure pictures, can will make vertical Gray Projection curve form a protruding peak in finder's face region with certain width, the border, the left and right sides at this protruding peak roughly is exactly the border, the left and right sides of people's face, this is that human face region often has higher brightness because compare with background;

At people's face left and right sides boundary, the summation of brightness value reduces rapidly on the vertical direction, thereby forms a tangible protruding peak;

Therefore, only need to determine the border, the left and right sides at main protruding peak in the vertical Gray Projection curve, can obtain the border, the left and right sides of people's face; In order to remove The noise, vertical Gray Projection curve is carried out smoothing processing with median filter method, the smoothing processing function is:

$\mathrm{PVF}\left(x\right)=\frac{1}{K+1}\underset{i=x-K/2}{\overset{x+K/2}{\mathrm{\&Sigma;}}}\mathrm{PV}\left(i\right)$

In the formula, K is the width of filter window, and its value is relevant with the size of people's face in image, and here, the K value is 6;

2) the human eye horizontal level estimates

After the border, the left and right sides of people's face was determined, the size of establishing border, the left and right sides was m, and then Ci Shi image size is m * N, with quadrat method to the horizontal Gray Projection function of this image-region was:

$\mathrm{PH}\left(x\right)=\underset{x=1}{\overset{m}{\mathrm{\&Sigma;}}}I(x,y)$

$\mathrm{PHF}\left(x\right)=\frac{1}{L+1}\underset{i=y-L/2}{\overset{y+L/2}{\mathrm{\&Sigma;}}}\mathrm{PH}\left(i\right)$

The value of L is relevant with the size of people's face in image, gets L=4 in the experiment;

3) estimation in human eye segmentation threshold interval and accurately location

Obtain the looks area I ₀After, only surplus eyebrow and the eyes that need be distinguished, can human eye therefrom be separated by seeking optimal segmenting threshold, so-called optimal segmenting threshold refers to can be with iris, pupil, go up the gray scale segmentation threshold that eye frame and eyebrow obviously separate, because the gray-scale value at iris, pupil and last eye frame position is obviously than its adjacent domain--circumference of eyes skin gray-scale value is low, therefore can cut apart iris, pupil and go up that the gray-scale value at eye frame position always exists, and be not unique, exist certain continuously among a small circle;

Estimate initial threshold T ₀Detect first peak dot of smoothed histogram, because the histogram after level and smooth two peak dots seldom occur in continuous 10 gray levels, so its preceding 250 gray levels all can be divided into 25 intervals, obtain the minimum and maximum value of 10 gray levels in each interval, search for the position of first peak dot by low gray area to high gray area then, estimate initial threshold T thus ₀

4) adjust threshold value automatically and determine eye position

T ₀Provide one the initial threshold that human eye can be split from image is arranged, but optimal segmenting threshold not necessarily in order to find optimal segmenting threshold, can be selected a step-length T _Step, determine optimal threshold by increasing step-length;

At first, use T ₀To image I ₀Binaryzation judges that whether the human eye black patch occurs, if black patch do not occur, adds a step-length T on the basis of last threshold value _Step, occur up to the eyes black patch, and black block in the binary image is carried out mark, calculate every area--pixel count, determine every occupied rectangle position and wide height;

Utilize above-mentioned segmentation threshold to increase progressively, detect and locate the whole process of eyes, wherein try to achieve initial threshold T ₀=60, get T in the experiment _Step=15, when threshold value T=105, the eyes black patch is detected, thereby reaches the accurate location of human eye.

3. the three-dimensional face modeling method based on positive silhouette as claimed in claim 1 or 2, it is characterized in that: described general faceform is generation according to specific human face characteristic point to specific faceform's conversion, mainly be to adopt based on the eye locating method of Gray Projection function and nose location algorithm the characteristic point position that the eyes and the nose of bidimensional image positions was carried out convergent-divergent to the integral grid size earlier in proportion with original sizing grid, be the reference point translation with the nose again, make the coordinate transform one-tenth of original mesh and two dimensional image pixel coordinate identical substantially;

Perhaps further utilize the information in the eye location algorithm that the basic grid that coincide is finely tuned, it is further coincide.

4. as the arbitrary described three-dimensional face modeling method of claim 1-3 based on positive silhouette, it is characterized in that: described the specifiable lattice model is carried out interpolation, generate meticulousr grid model, main method is based on the Delaunay triangulation technology of shape analysis; By the analysis of 184 dough sheets, make and form a large amount of points in each dough sheet again, use the Delaunay function among the MATLAB to carry out triangulation, unique point and dough sheet quantity after the interpolation are reached more than 30000, to obtain people's face grid of exquisiteness more;

The step of triangulation is as follows:

1) shape of calculating tri patch adopts diverse ways to carry out the subdivision operation then, at first calculates the longest edge E as front piece Ti ₁, vice-minister limit E ₂, minor face E ₃Length is respectively e _Il, e _I2And e _I3Calculate:

$r_1=\frac{e_{i1}}{e_{i2}},$ $r_2=\frac{e_{i2}}{e_{i3}},$ $r_3=\frac{e_{i1}}{e_{i3}}$

2) for three kinds of different shapes, take different processing modes:

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃In the time of≤H, H gets 1.3 in the experiment, this moment tri patch the subdivision algorithm, wherein, E, F and G point are respectively limit E ₁, E ₂And E ₃Mid point;

With E, F and G join among the non-unique point set NNF that subdivision produces as non-unique point, use as the non-unique point in the interpolation iterative process next time, and mark is with AC in markedt simultaneously, and BC and AB are the limit in other tri patchs on limit; The ABC that deletes patch in the dough sheet set adds dough sheet AEF, BEG, and FCG then;

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃When≤H can not satisfy, for the sake of simplicity, directly find longest edge E ₁Mid point E, judge then and work as r ₂〉=H ₁The time, then triangulation shown in accompanying drawing 3 (b), experimental selection H ₁=1.6 effects are best, and this moment, the F point was limit E ₂Mid point, with B, C, E, F spot projection utilize Delaunay trigonometric ratio algorithm then on XOY plane, return two tri patch BEC and ECF behind the trigonometric ratio;

Same non-unique point E and the F of adding adds the limit among dough sheet and the mark markedt;

As 1.0≤r ₁≤ H ∩ 1.0≤r ₂≤ H ∩ 1.0≤r ₃≤ H can not satisfy, and r ₂＜H ₁The time, illustrate that tri patch is longer and narrower this moment, finds longest edge E ₁Mid point E, then the E point is added among the NNF as non-unique point; Be the limit of adding among dough sheet and the mark markedt then;

3) concrete subdivision process is as follows: select each the tri patch Ti among the Pi_tris successively, at first in the first round processing, temporarily do not consider two kinds of tri patchs, the subdivision first round is handled all tri patchs with shape shown in the accompanying drawing 3c;

Judge the mark situation on Ti each limit in markedt,, then use method subdivision shown in the accompanying drawing 3a certainly if minor face or vice-minister limit have been labeled; If do not have the limit to be labeled or longest edge is labeled, then use method subdivision shown in the accompanying drawing 3c; Second takes turns in the processing, considers the tri patch of two kinds of shapes shown in accompanying drawing 3a and the accompanying drawing 3b;

Equally, if minor face is labeled, then use method subdivision shown in the accompanying drawing 3a;

If minor face is not labeled, then look a certain method subdivision of the concrete shape of tri patch in selecting shown in accompanying drawing 3a or the accompanying drawing 4b;

5. as the arbitrary described three-dimensional face modeling method of claim 1-4 based on positive silhouette, it is characterized in that: the pixel of described use X-Y scheme is carried out pinup picture to grid, and carry out texture, the color of the pixel of the bidimensional image that employing is corresponding is attached to the point on the grid, pass through bilinear interpolation--Bilinear Interpolation technology is carried out texture, determine earlier 4 pixels near pixel, adopt the difference that reduces the texture scale block diagram to mend then and calculate, its result determines the color of this point.

6. as the arbitrary described three-dimensional face modeling method based on positive silhouette of claim 1-5, it is characterized in that: the step that described texture is carried out texture mapping may be summarized to be:

Definition texture image, control texture, explanation texture mapping mode, definition texture coordinate;

The definition texture image comprises form, the texture image data type of the length of resolution, texture image of given texture image and width, texture image, the memory address of texture image;

Mapping between the coordinate of employing cylindrical surface projecting realization three-dimensional face model and the position of 2 d texture picture.

Images