KR20010084670A - Texture mapping method and apparatus for 2D facial image to 3D facial model - Google Patents

Abstract

PURPOSE: A method and apparatus is provided, in which control point and related vertexes are processed in an individual unit, to thereby reduce data process speed and obtain natural texture. CONSTITUTION: A method comprises the first step(200-220) of grouping pre-defined control points reflecting characteristics of a face in a three-dimensional face model, into individual unit of face component, and adjusting, in individual units, control points to be near to a two-dimensional face image; the second step(230-250) of calculating different displacement vectors in each unit with respect to vertexes contained in the unit except the control units, and obtaining a transformed three-dimensional face model; and the third step(260,270) creating texture in three-dimensional coordinate system of the two-dimensional face image by using coordinates of the transformed three-dimensional face model.

Description

Texture mapping method and apparatus for 2D facial image to 3D facial model}

The present invention relates to a method and apparatus for generating a 3D image, and more particularly, to a texture mapping method and a device from a 2D face image to a 3D face model.

Face modeling is also a difficult part in 3D modeling. This is because the face is very curved, and it is a part where a person can easily see a difference even with subtle changes. Personally, even if a standard model is created and used, the deformation methods required for each part are different. Therefore, a lot of experience and time are required for modeling and animation using a generalized 3D graphic program.

The generation of the face model can be viewed in two directions depending on the application. First, as a high quality face model used in movies, various 3D modeling tools are gradually including functions for face modeling. The face model generated by such a tool is sophisticated, but there is a problem in real time processing because it uses a general model transformation method for implementing animation. For this reason, some animation production companies have developed their own programs for face modeling privately. Second, a simplified face model used in games, etc., with a focus on real-time animation. In both cases, however, during the modeling of the face, many steps must be manually adjusted, and the quality of the face model generated by the modeler's capabilities and the additional changes that occur during the animation will vary. Therefore, for efficient animation, a simple face modeling method and associated facial animation tools are required.

The face model generation method using the conventional modeling software is divided into two methods: scratch building and template correction. Scratch Building is a new way to create face models using tool-provided polygon patches or Bezier patches. On the other hand, the template modification is to make a model by modifying the already produced template, which is unnatural when the model is used for later animation without the standardized model has undergone sufficient review. In addition, since the vertices constituting the template must be corrected directly in the process of modifying the template, it is difficult to generate a natural model. Another general purpose method is 3D scanning of the model, which includes an excessive increase in the number of polygons and difficulty in linking with animation.

In the prior art, a method of generating a face model by modification of a standard model is currently used by various research institutes, and is used in a slightly different method. In most cases, the basic concept of setting a control point and adjusting the control point to transform the standard model is the same. However, the position and number of control points are different, and the interpolation functions for the vertices other than the control point are different. Among them, a technique of blending a boundary between a portion where a front image and a side image are connected is commonly used as a texture processing method.

However, the model generation method by the modification of the conventional standard model has the following disadvantages. First, since the control points are not tied together as objects, the correlation between the control points is not taken into consideration, and accordingly, the control points need to be adjusted individually.

Second, because only one method is applied to the entire model, the vertex distribution characteristics of each part are ignored. If the distribution of vertices in the model is constant, one method can be considered.In general, however, the number of vertices is assigned to areas with high bends, such as the nose and lips, and the number of vertices is assigned to areas with less bends, such as cheeks. . Thus, when one method is applied, the magnitude of the calculated displacement vector of the vertex is too large or too small. Simple linear interpolation over the whole does not reflect features when creating a curriculum or unique face.

Third, even in the case of creating or automating texture generation manually using a conventional image processing tool in a conventionally developed system, since only the boundary of the texture is blended, there is a sense of heterogeneity of the image resulting from the absolute brightness difference between the front and side images. Can't solve it. In other words, when the illumination is not compensated, the ear part of the synthetic face seen from the front becomes brighter or darker than the actual image, thereby reducing the realism.

SUMMARY OF THE INVENTION The present invention provides a texture mapping method of a two-dimensional face image to a three-dimensional face model, which generates a natural texture while processing a control point and related vertices on a per-object basis, thereby easily and quickly processing a user interface. To provide.

Another object of the present invention is to provide a texture mapping apparatus of a two-dimensional face image to a three-dimensional face model performing the texture mapping method.

1 is a block diagram of a texture mapping apparatus of a two-dimensional face image to a three-dimensional face model according to the present invention.

2 is a flowchart illustrating a texture mapping method of a two-dimensional face image to a three-dimensional face model according to the present invention.

3 is a detailed block diagram according to a preferred embodiment of the individual adjustment unit shown in FIG. 1.

FIG. 4 is a detailed block diagram according to a preferred embodiment of the entire displacement vector calculator shown in FIG. 1.

5 is a detailed block diagram of a preferred embodiment of the texture preprocessor shown in FIG. 1.

6 is a view showing an example of a three-dimensional face model and its modified model used in the present invention.

7 is a diagram illustrating an example of texture mapping according to the present invention obtained through an experiment.

In order to achieve the above object, the texture mapping method of a two-dimensional face image to a three-dimensional face model according to the present invention,

(a) treating the predefined control points reflecting the feature points of the face in the three-dimensional face model as individual objects by combining each component of the face, and adjusting the control points on an individual basis to approach the input two-dimensional face image; b) calculating different displacement vectors within the object for vertices included in objects other than control points to obtain a modified three-dimensional face model and (c) two-dimensional using coordinates of the modified three-dimensional face model. Generating a texture in a three-dimensional coordinate system of the face image.

In order to achieve the above object, the texture mapping apparatus of the two-dimensional face image to the three-dimensional face model according to the present invention,

Pre-defined control points reflecting the feature points of the face in the three-dimensional face model are treated as objects by grouping each component of the face, and the individual unit three-dimensional face that roughly adjusts the control points in units of objects according to the input two-dimensional face image. The model transformation unit, which adjusts the control points in the object in detail, calculates the displacement vector of each control point, the final deformation by calculating different displacement vectors in the object for the vertices included in the object other than the control points It includes a global displacement vector calculation unit for obtaining a three-dimensional face model and a texture generation unit for generating a texture in the three-dimensional coordinate system of the two-dimensional face image using the coordinates of the modified standard model.

Hereinafter, with reference to the accompanying drawings, a method and apparatus for mapping a texture to a three-dimensional face model of a two-dimensional face image according to the present invention will be described as follows.

The present invention relates to a method of inputting a human face image as a two-dimensional digital image, extracting a mapping point for use in three-dimensional animation by semi-automatic texture mapping into a three-dimensional face model, and generating a natural texture. Natural facial animation requires the creation of natural facial models and sophisticated texture matching. The present invention utilizes a method of transforming a standard model to generate a face model, and also uses a method for extracting mapping points for natural texture matching and naturally connecting textures.

1 is a block diagram of a texture mapping apparatus of a two-dimensional face image to a three-dimensional face model according to the present invention, wherein the individual unit three-dimensional face model deformation unit 100, the object adjusting unit 110, and the total displacement vector calculation unit ( 120, a texture preprocessor 130, a cylindrical coordinate system texture generator 140, and a data storage 150.

2 is a flowchart illustrating a texture mapping method of a two-dimensional face image to a three-dimensional face model according to the present invention.

Specifically, referring to FIGS. 1 and 2, the object-based three-dimensional face model deformation unit 100 is first photographed on at least the front and side surfaces of a face through a video camcorder, various camera input devices, a scanner, an image file, and the like. A two-dimensional face image including two images of a front face image and a side face image is input (operation 200).

The individual unit 3D face model deforming unit 100 treats the predetermined control points reflecting the feature points of the face in the predefined 3D face model for each component of the face as an object. The predefined three-dimensional face model is modified by using two or more face images. In the present invention, the control points are individually adjusted in units of objects in the three-dimensional face model (step 210). For each object adjusted in step 210, a series of object numbers, positions, sizes, and displacements of rotation angles are obtained (step 220).

In general, the predefined three-dimensional face model, that is, the standard model, is indexed for all points constituting the model for later animation. Thus, you know in advance which point should be located on your face. The point to be used as a control point among the points is selected as a point that appears according to the boundary line extraction of the face contour in the face image, that is, the point existing on the boundary line. The control points selected in the present invention are handled as individual objects by tying, for example, eyes, noses, mouths, ears, etc. for each component of the face, and are adjusted to approximate a two-dimensional face image in units of objects. That is, the deformation unit of the three-dimensional face model becomes an object instead of individual control points, and the three-dimensional face model is roughly adjusted through the movement, rotation, and size deformation of the object.

Here, each object has information about the serial number, the center position, the size and the rotation angle. In addition, each entity has index information of control points. Through the user interface, the control points of the 3D face model are roughly adjusted in units of objects in accordance with the 2D face image. For example, after a user manipulates a predefined three-dimensional face model to overlap a two-dimensional face image displayed on a computer screen, an arbitrary object is selected through an input device such as a mouse, and the position, size, and rotation angle thereof. By changing the back, the control points contained within the object are globally adjusted.

The adjustment of individual units provides a very intuitive interface to the user and can be used very effectively later on in the transfer to automatic matching technology. The present invention also allows each control point to adjust its position separately in order to increase the convenience and accuracy of detailed adjustments for the three-dimensional face model adjusted on an individual basis.

Next, the object adjusting unit 110 inputs a series of object numbers, positions, sizes, and displacements of rotation angles for each object adjusted from the object unit three-dimensional face model deformation unit 100, and through the user interface. Adjust the control points in the object that require the selected fine adjustment and calculate the displacement vector of each control point in the object. The displacement vector of each object applied to all control points in the object is calculated using the displacement amount obtained in operation 220, and the displacement vector of the control point in the object to be adjusted in detail is calculated using the displacement amount (operation 230).

3 is a detailed block diagram according to a preferred embodiment of the entity adjusting unit 110 shown in FIG. 1, and includes an entity selector 302, an entity information adjuster 304, an object-specific displacement vector calculator 306, and a control point-specific displacement vector. A calculator 308 is provided.

The object selector 302 enters the selected control point via the user interface for further adjustment. Enter the control point index exactly. The object selector 302 searches for which object the input control point index is included in by referring to the lookup table that stores the control point index included in the object number, and selects the object. For example, if the control point index is index11 in FIG. 3, entity number 1 is selected.

The entity information adjuster 304 inputs the displacement amount with respect to the position, size, and rotation angle of the entity adjusted by the entity number and the entity unit three-dimensional face model deformation unit 100 exactly as the entity selected by the entity selector 302. The entity information regulator 304 updates the information of the selected entity with reference to the displacement amount. For example, FIG. 3 shows that the position, size, and rotation angle of the object No. 1 stored in the lookup table have been changed.

The object-specific displacement vector calculator 306 applies the values of the position, size, and rotation angle of the object updated by the object information adjuster 304 to all the control points included in the object, and calculates the object-specific displacement vector. do. Preferably, the displacement vector for each object is obtained by adding the displacement amount of the position to a value multiplied by the displacement amount of the rotation angle with respect to the intrinsic coordinates of the three-dimensional face model. The intrinsic coordinates of the model are different for each of the control points, but the remaining values apply for all control points.

The displacement-specific displacement vector calculator 308 calculates the displacement-specific displacement vectors calculated by the object-specific displacement vector calculator 306 and the individual displacement vectors for the control points input to the object selector 302, that is, the user-adjusted control points. Add up to calculate the final displacement vector of the control point. In other words, the control point selected by the user is adjusted in units of objects on the intrinsic coordinates of the model, and then has individual displacement vectors according to the detailed adjusted results.

Referring again to FIG. 1, the total displacement vector calculator 120 uses the displacement vectors of the control points obtained by the object adjuster 110 to calculate different displacement vectors in the object with respect to vertices included in the object in addition to the control points. Calculate (step 240). In addition, in addition to the control points, the displacement vector for the vertices may be different from object to object. By calculating the displacement vector of the vertex in the global displacement vector calculation unit 120 to obtain a final modified three-dimensional model.

When the coordinates of the vertices other than the control points are automatically adjusted in the 3D face model, different displacement vectors are calculated for each object in order to minimize distortion and reduce computation time. As a preferred embodiment, each of the control points is weighted in proportion to the distance to the vertices in the object to be adjusted. The product of the displacement vector and the weight of each of the control points calculated in operation 230 is accumulated, and the displacement vector of the vertex is obtained by arithmetic averaging by the number of control points.

FIG. 4 is a detailed block diagram according to a preferred embodiment of the total displacement vector calculator 120 shown in FIG. 1, which includes a weight calculator 402, an accumulator 404, a first vertex-specific displacement vector calculator 406, and 2 vertex displacement vector calculator 408 and selector 410.

The weight calculator 402 sequentially calculates the distance between the vertex and the control points for a vertex in the object to be adjusted. The distance obtained is put into a normal distribution function with experimentally determined variance values, and the result is used as a weight to obtain a value between 0 and 1. An equation of weight = exp (−distance / dispersion) may be used, so that the closer the distance between the target vertex and the control point is, the closer to 1. That is, the weight calculator 402 causes the displacement vector of the control points close to the target vertex to further affect the target vertex.

The accumulator 404 accumulates the product of the displacement vectors of the control points obtained by the object adjusting unit 110 and the respective weights calculated by the weight calculator 402 for the corresponding control points. The first vertex-displacement vector calculator 406 divides and accumulates the value accumulated by the accumulator 404 by the sum of the weights of the control points, and the average value becomes the displacement vector of the vertex. In this way we compute the displacement vectors for the vertices except for the control points in the object.

The result of this displacement vector calculation method is not good in all areas of the face. To compensate for this, the second vertex displacement vector calculator 408 calculates an object displacement vector applied to all the control points based on the object displacement amount in the object adjuster 110, and thus determines the position of the vertex based on the object displacement amount. Find the displacement vector.

The selector 410 is basically a large value, i.e., the amount of change in the outputs of the first and second per-vertex displacement vector calculators 406 and 408, for an object with dense vertices among the objects, such as the nose and the lip area. Select this larger value and use it as the displacement vector of the vertex.

Referring back to FIG. 1, the texture preprocessor 130 compensates for the difference in illumination by selecting a specific area of a face sensitive to illumination with respect to the two-dimensional face image input to the texture mapping apparatus according to the present invention.

When the displacement vector of the vertex is calculated through the step 240, the displacement vector for all the points of the three-dimensional face model is obtained, and the modified three-dimensional face model is obtained. Registration coordinates are determined by the modified three-dimensional face model. At this time, the texture is preprocessed before mapping the texture of the 2D face image to the modified 3D face model (operation 250). That is, lighting compensation using image values of specific regions is introduced as a preprocessing step for a natural connection of a two-dimensional face image to be used as texture data.

In general, the input texture data of a two-dimensional face image has a brightness that is different from other lighting directions in most cases of the front and side surfaces. Compensation of the illumination direction is usually very difficult, but compensation of brightness can have a significant effect by selecting the appropriate reference area of the light-sensitive face.

FIG. 5 is a detailed block diagram according to a preferred embodiment of the texture preprocessor 130 shown in FIG. 1, with specific area color average calculators 502, 504, 506, ratio calculators 512, 514 and color compensators 522, 524. FIG. do.

The specific area color average calculators 502, 504 and 506 input the right image, the front image and the left image of the face from the two-dimensional face image to be used as the texture data, respectively, and input the full displacement vector calculator 120 to extract the specific region of the face. Use the coordinates of the specific vertex obtained from. Based on a specific vertex, a certain area around the location is extracted as a specific area. For example, in FIG. 5, the basic reference vertex is determined as the edge position of the ball, and corresponds to a case in which the reference vertices are one left and right. Due to the gentle slope, the area around the ball has a small difference in illumination in the front and side images. The lighting difference in this part can be interpreted as the lighting difference of the whole image. This is because the direction of the normal vector in the vicinity of the ball is inclined approximately 45 degrees from the vertical to the front and approximately 45 degrees to the side. As a result, if the deviation of the illumination direction is not large, similar reflectance is obtained for the front and side images. Therefore, the vicinity of the ball mainly affects the brightness difference of the entire image.

The specific area color average calculators 502, 504, and 506 obtain the average values of R, G, and B of the pixel values around the reference vertices for the front and side images, respectively, in order to reduce errors such as blemishes or skin spots. The mean value may be an arithmetic mean or a median mean. Although the area of the specific area that takes the average value is variable, in the present example, an area having a diameter of 5 pixels is used for an image having a face area of about 25% at a size of 512x512. Since the effects of lighting are much on the left and right of the face, use independent average values on each side. In case of one side image, the other side is mirror symmetrically used. Even when copying, the color is adjusted based on the brightness distribution of the front image, resulting in a certain level of texture quality.

The ratio calculators 512 and 514 use the R, G and B average values R, G and B obtained for the front image and the R, G and B average values R ', G', B 'and R respectively obtained in the two side images. Calculate the ratio for RGB by adjusting the overall color by multiplying all the pixels in both sides of the image by the ratios obtained from the corresponding ratio calculators 512,514, respectively. At this time, since the human skin is not the primary color, the calculated new RGB value rarely exceeds the allowable range even if there are lighting differences, but the maximum value is exceeded if the RGB value exceeds the allowable range so that no error occurs. Decide on

On the other hand, in order to adjust the color more smoothly, when a plurality of reference vertices other than the reference vertex selected in FIG. 5 are used, a plurality of specific regions are also extracted. The specific area color average calculators 502, 504 and 506 calculate respective color average values for a plurality of specific areas, and the ratio calculators 512 and 514 also calculate the color average values in the front image and the color average values in both side images for each specific area. Calculate each ratio. The texture preprocessor 130 further includes linear interpolators (not shown), and the linear interpolators linearly interpolate the ratios obtained from the corresponding ratio calculators 512 and 514 with respect to the vertical axis. The color compensators 522 and 524 adjust the color of both side images by multiplying the corresponding pixel by a different ratio according to the position of the vertical axis of both side images.

Finally, in FIG. 1, the cylindrical coordinate system texture generator 140 uses the coordinates of the deformed three-dimensional face model finally obtained through the global displacement vector calculator 120, and preferably the texture preprocessor 130. In step 260, a texture is generated in the cylindrical coordinate system of the pre-processed 2D face image. Specifically, the texture and registration coordinates are converted into cylindrical coordinates by using the front image, the two side images compensated by the texture preprocessor 130, and the coordinates of the modified three-dimensional face model, and the boundary between the textures is generally known. Use boundary blending techniques to remove boundaries. The texture and registration coordinates in the cylindrical coordinate system are stored in the data storage unit 150 for later animation or still image production (step 270).

6 is a view showing an example of a three-dimensional face model and its modified model used in the present invention, (a) is the front and side of the three-dimensional face model expressed in units of units, (b) nose objects as individuals If you select to change the size of the nose object, it represents the front and side.

When the user selects a nose object and changes its size, the control points contained within the object are automatically adjusted globally. Next, when the fine adjustment of the control points was completed, vertices other than the control points contained in the object were also automatically adjusted to different displacement vectors.

7 is a diagram illustrating an example of texture mapping according to the present invention obtained through experiments, in which (a) illustrates a three-dimensional face model (indicated by a dashed line) prepared in advance to approach the front and side images of the input two-dimensional face image. (B) shows the generated model and (c) shows the generated texture, respectively. Referring to FIG. 7, it can be seen that the desired model and texture are easily generated only by adjusting the position and changing the size of the object in the 3D face model.

As described above, in the present invention, first, since the control points are grouped and used, correlations between the control points are taken into consideration, and by adjusting the control points on a per-unit basis, intuitive manual matching is possible, and an automatic matching technique is achieved by individualization. Facilitates the introduction of. Currently available feature extraction techniques basically start by finding the eye position and extract each feature point through the placement and size ratio of the face components that are already known. Therefore, the information obtained at each step of the feature point extraction technique can be directly applied to the object, and conversely, the correlation of control points can be applied to the feature point extraction process, thereby achieving a highly efficient system.

Second, in calculating the displacement vector, a calculation function capable of utilizing features for each face part is applied, thereby obtaining a more natural face model. Third, by preprocessing the texture color, the quality of the texture can be improved even if the image is obtained under a good lighting condition as well as the image obtained under a slightly different lighting condition.

Claims (17)

A method of mapping a two-dimensional face image to a three-dimensional face model by modifying a predefined three-dimensional face model with respect to an input two-dimensional face image and texture mapping the two-dimensional face image to a modified three-dimensional face model. To (a) treating predefined control points reflecting the feature points of the face in the three-dimensional face model as individual objects by binding them to each component of the face and adjusting them to approach the two-dimensional face image in units of objects; (b) calculating different displacement vectors in the object for vertices included in the object other than the control points to obtain a modified three-dimensional face model; And (c) generating a texture in the three-dimensional coordinate system of the two-dimensional face image by using the coordinates of the modified three-dimensional face model. Mapping method. According to claim 1, wherein the step (a), (a1) roughly adjusting the control points in the three-dimensional face model according to the two-dimensional face image on an individual basis, and obtaining a displacement amount for each adjusted object; (a2) selecting a control point requiring detailed adjustment and updating the information of the object including the control point with reference to the displacement amount obtained in step (a1); (a3) applying the updated information of the object to all control points included in the object, and calculating a displacement vector for each object; And (a4) calculating the final displacement vector of the control point included in the object by adding the individual displacement vector of the control point selected in the step (a3) to the individual displacement vector. Texture mapping method to dimensional face model. The method of claim 2, wherein the displacement amount is at least a displacement amount relative to the position, size, and rotation angle of the object, The displacement vector for each object is a texture obtained by adding a displacement amount of a position to a value obtained by multiplying the displacement amount of the rotation angle with respect to the intrinsic coordinates of the three-dimensional face model by the displacement amount of the size, and the texture to the three-dimensional face model. Mapping method. According to claim 1, wherein step (b), (b1) weighting each of the control points in proportion to a distance from a vertex in the object to be adjusted; And (b2) accumulating a product of a displacement vector of each of the control points and a corresponding weight based on the individual displacement amount adjusted in step (a), and calculating a displacement vector of a vertex by arithmetically averaging the number of the control points; Texture mapping method of the two-dimensional face image to the three-dimensional face model comprising a. The method of claim 4, wherein step (b) comprises: (b3) obtaining a displacement vector of a vertex based on the displacement amount of each object; And (b4) selecting a larger value of the displacement vectors of the vertices obtained in the step (b2) and the step (b3) for the object having the denser vertices. Texture mapping method of image to 3D face model. The method according to claim 1, wherein a predetermined region of a face sensitive to illumination is selected for the input two-dimensional face image to compensate for the difference in illumination, and the texture of the pre-processed two-dimensional face image is used in step (c). A texture mapping method of a two-dimensional face image to a three-dimensional face model, characterized in that. The method of claim 6, wherein the pretreatment step, Using the coordinates of the modified three-dimensional face model obtained in step (b), a specific region of the front face image and both side face images of the two-dimensional face image to be used as a texture is determined, and each color of the specific region is determined. Calculating average values; Obtaining respective ratios of the color average value in the front face image and the color average value in both side face images; And And adjusting the color tone of the two side images by multiplying the ratios by all the pixels of the both side face images. The method of claim 6, wherein the pretreatment step, By using the coordinates of the modified three-dimensional face model obtained in the step (b), a plurality of specific regions of the front face image and both side face images for the two-dimensional face image to be used as a texture is determined, and the plurality of specific regions Calculating respective color average values for; Obtaining ratios of the color average value of the front face image and the color average value of both side face images for each specific region, respectively; And Linearly interpolating the obtained ratios with respect to the vertical axis, and adjusting the color tone of the two side images by multiplying the corresponding pixels by different ratios according to the position of the vertical axis of the both side face images. How to map textures to models. A texture mapping apparatus of a two-dimensional face image to a three-dimensional face model, which deforms a predefined three-dimensional face model with respect to an input two-dimensional face image, and texture maps the two-dimensional face image to a modified three-dimensional face model. To In the three-dimensional face model, an object unit that treats predefined control points reflecting feature points of a face as an object by combining each component of the face and adjusts the control points in units of objects approximately according to the input two-dimensional face image. Dimensional face model deformation; An entity adjusting unit that adjusts the control points in the object in detail and calculates a displacement vector of each control point; A total displacement vector calculator configured to calculate different displacement vectors in the object for vertices included in objects other than the control points to obtain a final modified three-dimensional face model; And And a texture generator configured to generate a texture in a three-dimensional coordinate system of the two-dimensional face image by using the coordinates of the modified standard model. The method of claim 9, wherein the individual adjustment unit, An object selector for selecting the object including the control point selected by the user for fine adjustment; An entity information adjuster for updating the information of the selected entity with reference to the displacement amount obtained for each entity adjusted by the entity unit three-dimensional face model deformation unit; An object-specific displacement vector calculator which applies the updated object information to all the control points included in the object and calculates the object-specific displacement vector; And And a control point displacement vector calculator for calculating the final displacement vector of the control point included in the object by adding the individual displacement vectors of the selected control points to the object-specific displacement vectors. Texture mapping device. 11. The method of claim 10, wherein the displacement amount is at least a displacement amount relative to the position, size and rotation angle of the object, The displacement vector for each object is a texture obtained by adding a displacement amount of a position to a value obtained by multiplying the displacement amount of the rotation angle with respect to the intrinsic coordinates of the three-dimensional face model by the displacement amount of the size, and the texture to the three-dimensional face model. Mapping device. The method of claim 9, wherein the total displacement vector calculation unit, A weight calculator for calculating a weight for each of the control points in proportion to the distance to the vertex in the object to be adjusted; An accumulator for accumulating a product of the displacement vector of each of the control points based on the object-to-object displacement adjusted by the individual unit three-dimensional face model deformation unit and the respective weights calculated by the weight calculator; And And a first vertex displacement vector calculator for calculating a displacement vector of a vertex by arithmetically averaging an accumulated value based on the number of control points. The method of claim 12, wherein the displacement vector calculation unit, A second vertex displacement vector calculator for calculating a displacement vector of a vertex based on the displacement amount of each object; And A three-dimensional face model of the two-dimensional face image further comprising a selector for selecting a larger value among the outputs of the first and second per-vertex displacement vector calculators for the dense vertices among the objects. Texture mapping device. The texture mapping apparatus of claim 9, Comprising a specific area of the light-sensitive face with respect to the input two-dimensional face image to compensate for the difference, and further comprises a texture pre-processing unit for outputting the compensated two-dimensional face image to the texture generation unit Texture mapping device for 3D face model of face image. The method of claim 14, wherein the texture preprocessor, First to the first region to determine the specific region of the front face image and both side face image for the input two-dimensional face image by using the coordinates of the modified three-dimensional face model, and calculate the respective color average values for the specific region A third color average calculator; First and second ratio calculators for calculating respective ratios of a color average value in the front face image and a color average value in both side face images; And first and second color compensators for adjusting the colors of both side face images by multiplying the ratios by all the pixels of both side images. The method of claim 14, wherein the texture preprocessor, A first method for determining a plurality of specific regions of a front face image and a side face image of an input two-dimensional face image by using the coordinates of the modified three-dimensional face model, and calculating respective color average values for the specific region. To third color average calculator; First and second ratio calculators for calculating respective ratios of the color average value in the front face image and the color average value in both side images for each specific region; First and second linear interpolators for linearly interpolating the ratios relative to a longitudinal axis; And And a first and second color compensators for adjusting the colors of both side face images by multiplying the corresponding pixels by different ratios according to the vertical axis positions of both side images. Device. The method of claim 9, wherein the three-dimensional coordinate system, A texture mapping device of a two-dimensional face image to a three-dimensional face model, characterized in that the cylindrical coordinate system.