CN108986221A - A kind of three-dimensional face grid texture method lack of standardization approached based on template face - Google Patents
A kind of three-dimensional face grid texture method lack of standardization approached based on template face Download PDFInfo
- Publication number
- CN108986221A CN108986221A CN201810841353.2A CN201810841353A CN108986221A CN 108986221 A CN108986221 A CN 108986221A CN 201810841353 A CN201810841353 A CN 201810841353A CN 108986221 A CN108986221 A CN 108986221A
- Authority
- CN
- China
- Prior art keywords
- template
- dimensional face
- face grid
- texture
- grid
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/04—Texture mapping
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Graphics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Processing Or Creating Images (AREA)
- Image Generation (AREA)
Abstract
The invention discloses a kind of three-dimensional face grid texture methods lack of standardization approached based on template face, comprising the following steps: establishes template three-dimensional face grid to the texture mapping relationship between template texture maps picture point;Template three-dimensional face grid is tentatively aligned with three-dimensional face grid lack of standardization;The error energy function between template three-dimensional face grid and three-dimensional face grid lack of standardization after building alignment, and optimize the error energy function, so that template three-dimensional face grid approaches three-dimensional face grid lack of standardization, finally obtain specification three-dimensional face grid;According to above-mentioned texture mapping relationship, the specification three-dimensional face grid of generation and the texture mapping of template texture image are realized, to generate the textured three-dimensional face model of specification.Specification three-dimensional face grid of the invention realizes the texture mapping with template texture image by template parameter texture coordinate, has the advantage of specification input three-dimensional face network topology and vertex number, low cost.
Description
Technical field
The present invention relates to human face data computer processing technical field more particularly to a kind of approached not based on template face
Specification three-dimensional face grid texture method.
Background technique
Textured three-dimensional face model is widely used in personalized consumption, film advertisement, Entertainment, visualization in real time
The numerous areas such as communication, cosmetic surgery preview, human-computer interaction of new generation.In addition, texture mapping be establish three-dimensional object surface and
The process that two dimensional image space pixel coordinate pair is answered.Less hardware computing resource and rendering more conveniently spy are consumed because having
Point, so texture mapping is the effective ways for enhancing the three-dimensional face grid sense of reality.But in many applications, pass through some sides
The problem that the generally existing topology of three-dimensional face grid of formula building is lack of standardization or vertex number is larger, so that establishing this three manually
The process for tieing up face grid texture mapping spends a large amount of manpowers and time cost.Meanwhile this three-dimensional face grid is more difficult is counting
The poor terminal platform of calculation ability (such as smart phone) carries out smooth rendering.Therefore, it is a kind of can automatically establish it is lack of standardization
Three-dimensional face grid texture mapping simultaneously can optimize the method for network topology and vertex number with biggish application value.
In the texture-mapping algorithm of three-dimensional face grid, found by the literature search to the prior art, Blanz et al.
(BlanzV,Vetter T.A morphable model for the synthesis of 3D faces[C]//
Proceedings of the 26th annual conference on Computer graphics and
interactive techniques.ACM Press/Addison-Wesley Publishing Co.,1999:187-194.)
It is proposed can deformation model (3D Morphable Model, 3DMM) carry out three-dimensional face reconstruction and texture mapping method, lead to
Cross perspective projection and particular light algorithm can deformation model be rendered to two-dimension human face image, by optimizing the two dimensional image and defeated
The pixel square error for entering image, obtain can deformation model texture image Combining weights.But it can deformation needed for this method
Model is statisticallyd analyze dependent on large-scale three dimensional face database, and Given Face database sharing, maintenance and model training need
Expend a large amount of manpower financial capacities, and the parameter space due to being limited to database, human face rebuilding result restricted application.Than
Such as, the face prediction model trained using European face database is difficult to reconstruct preferable Asia face.Open sword etc.
(sword fusion SFM and the video flowing three-dimensional expression of dynamic texture mapping rebuild [J] CAD and graphics
Report, 2010,22 (6): 000949-958.) propose it is a kind of fusion SFM and dynamic texture mapping method, will tracking obtain it is every
Frame two dimensional character point position coordinates regard the texture coordinate on predefined one group of feature vertex on three-dimensional face model as, thus logical
It crosses interpolation and the face texture information extracted from original video is mapped into reconstruction corresponding with each video frame frame by frame automatically
Faceform out can generate three-dimensional face expression animation true to nature from the original face video that monocular cam is shot.But it should
Method is what the principle based on machine vision was rebuild, and due to lacking the priori knowledge of face, therefore the sense of reality of reconstruction model needs
It further increases.Method (Newcombe R A, Izadi S, the Hilliges O, et of Newcombe et al.
al.KinectFusion:Real-time dense surface mapping and tracking[C]//Mixed and
augmented reality(ISMAR),2011 10th IEEE international symposium on.IEEE,2011:
127-136.) depth camera Kinect can be used to carry out 3-D scanning and reconstruction to real human body, finally obtains textured information
3 D human body, but the topological more complex and vertex quantity of model meshes that this method is rebuild is also larger, need to pass through long period people
Work reparation just can apply to engineering project;And it is similar to there are problems that this using the method for spatial digitizer.
Summary of the invention
In view of the deficienciess of the prior art, it is an object of the present invention to provide a kind of lack of standardization three approached based on template face
Face grid texture method is tieed up, the present invention is three-dimensional by constructing the template with parametrization texture coordinate and texture image in advance
Face, and it is directed to the three-dimensional face grid lack of standardization of different inputs, algorithm calculating is carried out with the template face, to make this
Invent advantage at low cost, in addition there is the present invention specification to input three-dimensional face network topology and vertex number.
To achieve the goals above, the present invention is to realize by the following technical solutions:
A kind of three-dimensional face grid texture method lack of standardization approached based on template face of the invention, including with
Lower step:
1) template comprising template three-dimensional face grid, template parameter texture coordinate, template texture image is constructed
Three-dimensional face grid, and the template three-dimensional face grid is established to the texture mapping relationship between template texture maps picture point;
2) the template three-dimensional face grid is tentatively aligned with three-dimensional face grid lack of standardization;
3) error energy between the template three-dimensional face grid and three-dimensional face grid lack of standardization after building alignment
Function, and optimize the error energy function by gradient descent method and least square method, so that the template three-dimensional face
Grid approaches three-dimensional face grid lack of standardization, finally obtains specification three-dimensional face grid;
4) according to the texture mapping relationship of step 1), the specification three-dimensional face grid and template line generated in step 3) is realized
The texture mapping for managing image, to generate the textured three-dimensional face model of specification.
In step 1), the method for building up of the texture mapping relationship is as follows:
The first step has the template three-dimensional face grid of five features using 3 d modeling software manual construction one;
Second step, by cylindrical surface projecting mode, by template three-dimensional face Grid Projection to two-dimensional surface region [0,1]2I.e.
Texture space, and two dimension pattern plate parametrization texture coordinate is obtained, wherein grid and parametrization texture coordinate point exist maps one by one
Relationship, mapping relations can be obtained by following formula one by one:
f(pi)=qi
Wherein, piIt is the vertex on template three-dimensional face grid, qiIt is template parameter texture coordinate point;
Third step uses the manual structure of two dimensional image processing software according to the position distribution of template parameter texture coordinate point
A template texture image is built, wherein template parameter coordinate points and template texture maps picture point have mapping relations one by one, one by one
Mapping relations can be obtained by following formula:
g(qi)=pixeli
Wherein, pixeliIt is the pixel of template texture image;
4th step obtains template three-dimensional face grid vertex and template line according to the function passes relationship of above-mentioned f and g
The texture mapping relationship of image is managed, then texture mapping relationship is g [f (pi)]=pixeli。
It is described to be tentatively aligned that the specific method is as follows in step 2):
The first step respectively leads the vertex set of the template three-dimensional face grid and three-dimensional face grid lack of standardization
Constituent analysis, the local coordinate system matrix for obtaining the template three-dimensional face grid is M1, origin O1, the three-dimensional lack of standardization
The local coordinate system matrix of face grid is M2, origin O2;
Second step calculates M1To M2Transition matrix, i.e., according to M2=M1T obtains transition matrix T=M1 -1M2;
Third step, by each vertex p of the template three-dimensional face gridiTransform to three-dimensional face grid lack of standardization
Under coordinate system, transformation for mula is as follows:
pi←T·[pi+(O2-O1)]。
The mistake between the template three-dimensional face grid and three-dimensional face grid lack of standardization in step 3), after alignment
Poor energy function is defined as follows formula:
F (W, V)=w1fdistance(V)+w2fsmooth(V)
Wherein, V={ p1,p2,...,pn},It is the vertex set of template three-dimensional face grid, { cpi|1≤i≤m}
It is the vertex set of three-dimensional face grid lack of standardization, W=(w1,w2) it is weight coefficient, w1, w2Indicate the weight of two bound terms, m
Indicate the vertex number of three-dimensional face grid lack of standardization, cpiIndicate i-th of apex coordinate of three-dimensional face grid lack of standardization,It is distance restraint item, for measuring template three-dimensional face grid and three-dimensional face lack of standardization
Error between grid point cloud, Proji(V) indicate any one vertex of three-dimensional face grid lack of standardization to template three-dimensional face net
The nearest subpoint of lattice;It is smoothness constraint term, for ensure that template three-dimensional face grid
Deformed local smoothing method, diIndicate template three-dimensional face grid vertex piDegree, pijIndicate template three-dimensional face grid top
Point piJ-th of neighborhood point of single order.
In step 3), optimize the specific of error energy function using the gradient descent method and least square method
Method is as follows:
The first step initializes Wi=W0, Vi=V0;
Second step, fixed Wi, in ViOn the basis of construct least square linear equation group AVi+1=Bi+1, and solve and obtain
Vi+1, that is, pass through the template three-dimensional face grid of deformed NextState, wherein WiIndicate the weight coefficient of next state,
ViIndicate the template three-dimensional face grid vertex set of NextState, W0Indicate initial weight coefficient, V0Indicate initial template
Three-dimensional face grid vertex set;A indicates the relational matrix of bound term, Bi+1It is the relational matrix end value structure under NextState
At matrix;
Third step, fixed Vi+1, in WiOn the basis of the weight coefficient W of NextState is calculated with gradient descent methodi+1,
State renewal equation such as following formula:
Wherein, f ' (Wi) indicate energy function in WiUnder derivative, be a real number,Indicate real number, the step of gradient decline
Long value is α=0.00000001, due to WiItself is also a dual vector, therefore f function is to each weight real variable
Partial derivative calculate such as following formula:
And single wiChange step β take 0.000001;
4th step updates Vi←Vi+1, Wi←Wi+1, when meeting f ' (Wi)<ε1Or edistance<ε2Condition, termination optimized
Otherwise journey is continued to execute into second step, wherein relative distance error edistanceTo measure three-dimensional face grid vertex lack of standardization
Error between collection and template three-dimensional face grid, ε1Indicate f ' (Wi) upper limit value, ε2Indicate edistanceUpper limit value.
Above-mentioned relative distance error edistanceCalculation method such as following formula:
Wherein, fdistance(Vi) indicate that all vertex are nearest to template three-dimensional face grid on three-dimensional face grid lack of standardization
The square distance of subpoint and,Indicate all vertex on three-dimensional face grid lack of standardization to world coordinate system origin away from
From quadratic sum.
In step 1), there are three features for the template three-dimensional face grid tool:
1. template three-dimensional face network topology is symmetrical according to face's axis face;
2. the vertex number of template three-dimensional face grid, which less than 20,000, is subject to, is applicable in application and development;
3. the resolution ratio of template texture image is 1024*1024.
In step 3), there are three features for the specification three-dimensional face grid tool ultimately generated:
1. having five features similar with three-dimensional face grid lack of standardization;
2. having network topology identical with template three-dimensional face grid and vertex number;
3. sharing same template with template three-dimensional face grid parameterizes texture coordinate, can be carried out with template texture image
Texture mapping, mapping relations are still above-mentioned g [f (pi)]=pixeli。
The beneficial effects of the present invention are:
1. existing texture mapping method relies on the texture information in large-scale three dimensional face database or relies on expensive three-dimensional
Scanner obtains colour information, needs to expend a large amount of financial resources, manpower and time.And the present invention constructs a band parametrization in advance
The template three-dimensional face of texture coordinate and texture image, and it is directed to the three-dimensional face grid lack of standardization of different inputs, using should
Template face carries out algorithm calculating.Compared to existing method, the present invention is a kind of method of relatively low cost.
2. for three-dimensional face grid being obtained by spatial digitizer or manually constructing, there are network topologies
Lack of standardization, the biggish problem of vertex quantity, the present invention go to approach the three-dimensional face net lack of standardization by a template three-dimensional face
Lattice, ultimately generate and the three-dimensional face lack of standardization has similar five features but network topology specification and the moderate rule of vertex number
The textured three-dimensional face grid of the textured three-dimensional face grid of model, the specification can be directly used for secondary development, have relatively wide
General prospect of the application.
Detailed description of the invention
The dough sheet rendering effect of Fig. 1 (a) template three-dimensional face grid;
The wire frame rendering effect of Fig. 1 (b) template three-dimensional face;
Fig. 1 (c) template parameter texture coordinate;
Fig. 1 (d) template texture image;
The texture mapping effect picture of Fig. 1 (e) template three-dimensional face grid;
Fig. 2 is the algorithm schematic diagram for optimizing error energy function, and outer ring dotted line indicates three-dimensional face grid lack of standardization, internal
Solid line indicates template three-dimensional face grid;
The dough sheet rendering effect of Fig. 3 (a) three-dimensional face grid lack of standardization;
The wire frame rendering effect figure of Fig. 3 (b) three-dimensional face grid lack of standardization;
The dough sheet for the specification three-dimensional face grid that Fig. 3 (c) is ultimately generated renders;
The wire frame rendering effect for the specification three-dimensional face grid that Fig. 3 (d) is ultimately generated;
The texture mapping effect picture for the specification three-dimensional face that Fig. 3 (e) is ultimately generated;
Fig. 4 is method flow schematic diagram of the invention.
Specific embodiment
To be easy to understand the technical means, the creative features, the aims and the efficiencies achieved by the present invention, below with reference to
Specific embodiment, the present invention is further explained.
The present invention constructs a textured template three-dimensional face in advance and removes auxiliary human face rebuilding with this template face
Secondly process the priori knowledge of human face five-sense-organ is added in algorithm, most first without using face database to realize low cost
The standardization of model meshes topology and vertex quantity is also achieved eventually, and this method can be by nonstandard three-dimensional face Mesh Conversion
For network topology specification and the moderate specification face grid of vertex number, and realize and the texture mapping of texture image.
Referring to fig. 4, a kind of three-dimensional face gridding texture mapping method lack of standardization approached based on template face of the invention
Specific step is as follows:
1. building one specification comprising template three-dimensional face grid, template parameter texture coordinate, template texture image
Template face, and template three-dimensional face grid is established to the texture mapping relationship between template texture maps picture point.
The specific steps of the building of the template three-dimensional face grid are as follows:
The first step has the template three of five features using 3 d modeling software (3dmax, maya etc.) manual construction one
Face grid is tieed up, shown in dough sheet rendering effect such as Fig. 1 (a), shown in wire frame rendering effect figure such as Fig. 1 (b);
Second step, by cylindrical surface projecting mode, by template three-dimensional face Grid Projection to two-dimensional surface region [0,1]2(i.e.
Texture space), and two dimension pattern plate parametrization texture coordinate is obtained, wherein grid and parametrization texture coordinate point exist maps one by one
Relationship, the mapping relations one by one can be obtained by following formula:
f(pi)=qi
Wherein, piIt is the vertex on template three-dimensional face grid, qiIt is template parameter texture coordinate point;
The cylindrical surface projecting is using virgin crystalline substance in the thesis " void towards video display special efficacy in Shandong University in 2006
The method that 4.2.2 section (face mesh parameterization) proposes in quasi- performer's sense of reality three-dimensional face Modeling Research ", can be calculated
The parametrization texture coordinate of template three-dimensional face grid, as shown in Fig. 1 (c).
Third step uses two dimensional image processing software according to the position distribution of template parameter texture coordinate point
(Photoshop etc.) manual construction one opens template texture image, as shown in Fig. 1 (d);Wherein template parameter coordinate points and template
Texture maps picture point, which has mapping relations, the mapping relations one by one one by one, to be obtained by following formula:
g(qi)=pixeli
Wherein, pixeliIt is the pixel of template texture image;
4th step, according to the function passes relationship of above-mentioned f and g, available template three-dimensional face grid vertex and mould
The mapping relations one by one of plate texture image, the mapping relations one by one are g [f (pi)]=pixeli, as template three-dimensional face
Grid and template texture image texture mapping relationship, shown in texture mapping effect picture such as Fig. 1 (e).
There are three features for the template three-dimensional face tool of the specification: 1. template three-dimensional face network topology is according in face
Axial plane is symmetrical;2. the vertex number of template three-dimensional face grid, which less than 20,000, is subject to, is applicable in application and development;3. template texture maps
The resolution ratio of picture is 1024*1024.
2. by being tentatively aligned between template three-dimensional face grid and three-dimensional face grid lack of standardization.
The preliminary alignment, can be obtained by following steps:
The first step carries out principal component to the vertex set of template three-dimensional face grid and three-dimensional face grid lack of standardization respectively
Analysis, the local coordinate system matrix that can obtain template three-dimensional face grid is M1, origin O1, the office of three-dimensional face grid lack of standardization
Portion's coordinate system matrix is M2, origin O2;
The local coordinate system matrix M1And M2, it will be obtained by following procedure:
Firstly, utilizing bright thesis " the three-dimensional people based on Invariance feature pleased in 2013 in Beijing Jiaotong University
The principal component analytical method mentioned in face Study of recognition " 3.3.1 section (attitude updating of three-dimensional face model), can be calculated
Three major axes orientations and centroid position of three-dimensional face grid.Further, if the major axes orientation of template three-dimensional face grid is distinguished
For v1, v2, v3, thenIf the major axes orientation of three-dimensional face grid lack of standardization is respectively v1', v2', v3', then
Second step calculates and arrives M1To M2Transition matrix, i.e., according to M2=M1T, available transition matrix T=M1 -1M2;
Third step, by each vertex p of template three-dimensional face gridiTransform to the coordinate of three-dimensional face grid lack of standardization
Under system.Transformation for mula is as follows:
pi←T·[pi+(O2-O1)]
3. the error energy function between template three-dimensional face grid and three-dimensional face grid lack of standardization after building alignment,
And optimize the error energy function by gradient descent method and least square method, so that template three-dimensional face grid is approached and is not advised
Model three-dimensional face grid finally obtains specification three-dimensional face grid;
The error energy function between template three-dimensional face grid and three-dimensional face grid lack of standardization after the alignment
It is defined as follows formula:
F (W, V)=w1fdistance(V)+w2fsmooth(V)
Wherein, V={ p1,p2,...,pn},It is the vertex set of template three-dimensional face grid, { cpi|1≤i≤m}
It is the vertex set of three-dimensional face grid lack of standardization, W=(w1,w2) it is weight coefficient;It is
Distance restraint item, for measuring the error between template three-dimensional face grid and three-dimensional face grid point cloud lack of standardization, Proji
(V) nearest subpoint of expression any one vertex of three-dimensional face grid lack of standardization to template three-dimensional face grid;It is smoothness constraint term, for ensure that the part after template three-dimensional face distortion of the mesh is flat
Slip, diIndicate template three-dimensional face grid vertex piDegree, pijIndicate template three-dimensional face grid vertex piJ-th of single order adjacent
Domain point.
Fig. 2, described to optimize error energy function using gradient descent method and least square method specific step is as follows:
The first step initializes Wi=W0, Vi=V0;
Second step, fixed Wi, in ViOn the basis of construct least square linear equation group AVi+1=Bi+1, and solve and obtain
Vi+1, that is, pass through the template three-dimensional face grid of deformed NextState;
Third step, fixed Vi+1, in WiOn the basis of the weight coefficient W of NextState is calculated with gradient descent methodi+1,
State renewal equation such as following formula:
Wherein the step-length value of gradient decline is α=0.00000001.Due to WiIt itself is also a dual vector, therefore f
Function calculates such as following formula to the partial derivative of each weight real variable:
And single wiChange step β take 0.000001.
4th step updates Vi←Vi+1, Wi←Wi+1, when meeting f ' (Wi)<ε1Or edistance<ε2Condition, termination optimized
Otherwise journey is continued to execute into second step.
In 4th step, a kind of relative distance error e is defineddistanceTo measure three-dimensional face grid vertex lack of standardization
Error between collection and template three-dimensional face grid calculates such as following formula:
Wherein fdistance(Vi) indicate that all vertex are nearest to template three-dimensional face grid on three-dimensional face grid lack of standardization
The square distance of subpoint and,Illustrate on three-dimensional face grid lack of standardization all vertex to world coordinate system origin
Square distance and.By the algorithm schematic diagram of Fig. 2 it is found that outer ring dotted line indicates that three-dimensional face grid lack of standardization, inner ring solid line indicate
Template three-dimensional face grid, after error energy function optimization, template three-dimensional face grid can largely approach three-dimensional lack of standardization
Face grid illustrates that template three-dimensional face grid and three-dimensional face grid lack of standardization at this time have similar five features.
There are three features for the specification three-dimensional face grid tool ultimately generated: 1. having and three-dimensional face net lack of standardization
The similar five features of lattice;2. having network topology identical with template three-dimensional face grid and vertex number;3. with template three
It ties up face grid and shares same template parametrization texture coordinate, texture mapping, mapping relations can be carried out with template texture image
It is still above-mentioned g [f (pi)]=pixeli.Compare the three-dimensional face net lack of standardization it is found that before optimizing by Fig. 3 (a)-Fig. 3 (d)
The mixed and disorderly simultaneously vertex number of LF Topology is very big, and optimized three-dimensional face grid is with the relatively network topology of specification and suitably
Vertex number.
4. realizing the specification three-dimensional face and template texture image generated in step 3 according to the texture mapping relationship of step 1
Texture mapping, so that final textured three-dimensional face model is generated, as shown in Fig. 3 (e).
In conclusion the present invention has following innovation:
1. existing texture mapping method relies on the texture information in large-scale three dimensional face database or relies on expensive three-dimensional
Scanner obtains colour information, needs to expend a large amount of financial resources, manpower and time.And the present invention constructs a band parametrization in advance
The template three-dimensional face of texture coordinate and texture image, and it is directed to the three-dimensional face grid lack of standardization of different inputs, using should
Template face carries out algorithm calculating.Compared to existing method, the present invention is a kind of method of relatively low cost.
2. for three-dimensional face grid being obtained by spatial digitizer or manually constructing, there are network topologies
Lack of standardization, the biggish problem of vertex quantity, the present invention go to approach the three-dimensional face net lack of standardization by a template three-dimensional face
Lattice, ultimately generate and the three-dimensional face lack of standardization has similar five features but network topology specification and the moderate rule of vertex number
The textured three-dimensional face grid of the textured three-dimensional face grid of model, the specification can be directly used for secondary development, have relatively wide
General prospect of the application.
The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention.The technology of the industry
Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and the above embodiments and description only describe this
The principle of invention, without departing from the spirit and scope of the present invention, various changes and improvements may be made to the invention, these changes
Change and improvement all fall within the protetion scope of the claimed invention.The claimed scope of the invention by appended claims and its
Equivalent thereof.
Claims (8)
1. a kind of three-dimensional face grid texture method lack of standardization approached based on template face, which is characterized in that including following step
It is rapid:
1) it is three-dimensional to construct the template comprising template three-dimensional face grid, template parameter texture coordinate, template texture image
Face grid, and the template three-dimensional face grid is established to the texture mapping relationship between template texture maps picture point;
2) the template three-dimensional face grid is tentatively aligned with three-dimensional face grid lack of standardization;
3) the error energy function between the template three-dimensional face grid and three-dimensional face grid lack of standardization after building alignment,
And optimize the error energy function by gradient descent method and least square method, so that the template three-dimensional face grid is forced
Nearly three-dimensional face grid lack of standardization, finally obtains specification three-dimensional face grid;
4) according to the texture mapping relationship of step 1), the specification three-dimensional face grid generated in step 3) and template texture maps are realized
The texture mapping of picture, to generate the textured three-dimensional face model of specification.
2. the three-dimensional face grid texture method lack of standardization approached as described in claim 1 based on template face, feature are existed
In in step 1), the method for building up of the texture mapping relationship is as follows:
The first step has the template three-dimensional face grid of five features using 3 d modeling software manual construction one;
Second step, by cylindrical surface projecting mode, by template three-dimensional face Grid Projection to two-dimensional surface region [0,1]2I.e. texture is empty
Between, and two dimension pattern plate parametrization texture coordinate is obtained, wherein there are mapping relations one by one in grid and parametrization texture coordinate point, and one
One mapping relations can be obtained by following formula:
f(pi)=qi
Wherein, piIt is the vertex on template three-dimensional face grid, qiIt is template parameter texture coordinate point;
Third step uses two dimensional image processing software manual construction one according to the position distribution of template parameter texture coordinate point
Template texture image is opened, wherein template parameter coordinate points and template texture maps picture point have mapping relations one by one, map one by one
Relationship can be obtained by following formula:
g(qi)=pixeli
Wherein, pixeliIt is the pixel of template texture image;
4th step obtains template three-dimensional face grid vertex and template texture maps according to the function passes relationship of above-mentioned f and g
The texture mapping relationship of picture, then texture mapping relationship is g [f (pi)]=pixeli。
3. the three-dimensional face grid texture method lack of standardization approached as claimed in claim 2 based on template face, feature are existed
In described to be tentatively aligned that the specific method is as follows in step 2):
The first step carries out principal component to the vertex set of the template three-dimensional face grid and three-dimensional face grid lack of standardization respectively
Analysis, the local coordinate system matrix for obtaining the template three-dimensional face grid is M1, origin O1, the three-dimensional face lack of standardization
The local coordinate system matrix of grid is M2, origin O2;
Second step calculates M1To M2Transition matrix, i.e., according to M2=M1T obtains transition matrix T=M1 -1M2;
Third step, by each vertex p of the template three-dimensional face gridiTransform to the coordinate of three-dimensional face grid lack of standardization
Under system, transformation for mula is as follows:
pi←T·[pi+(O2-O1)]。
4. the three-dimensional face grid texture method lack of standardization approached as described in claim 1 based on template face, feature are existed
In the error energy between the template three-dimensional face grid and three-dimensional face grid lack of standardization in step 3), after alignment
Flow function is defined as follows formula:
F (W, V)=w1fdistance(V)+w2fsmooth(V)
Wherein,It is the vertex set of template three-dimensional face grid, { cpi| 1≤i≤m } it is lack of standardization
The vertex set of three-dimensional face grid, W=(w1,w2) it is weight coefficient, w1, w2Indicate the weight of two bound terms, m expression is not advised
The vertex number of model three-dimensional face grid, cpiIndicate i-th of apex coordinate of three-dimensional face grid lack of standardization,It is distance restraint item, for measuring template three-dimensional face grid and three-dimensional face lack of standardization
Error between grid point cloud, Proji(V) indicate any one vertex of three-dimensional face grid lack of standardization to template three-dimensional face net
The nearest subpoint of lattice;
It is smoothness constraint term, for ensure that the part after template three-dimensional face distortion of the mesh
Flatness, diIndicate template three-dimensional face grid vertex piDegree, pijIndicate template three-dimensional face grid vertex piJ-th of single order
Neighborhood point.
5. the three-dimensional face grid texture method lack of standardization approached as described in claim 1 based on template face, feature are existed
In in step 3), optimizing error energy function using the gradient descent method and least square method, the specific method is as follows:
The first step initializes Wi=W0, Vi=V0;
Second step, fixed Wi, in ViOn the basis of construct least square linear equation group AVi+1=Bi+1, and solve and obtain Vi+1, i.e.,
By the template three-dimensional face grid of deformed NextState, wherein WiIndicate the weight coefficient of next state, ViIt indicates
The template three-dimensional face grid vertex set of NextState, W0Indicate initial weight coefficient, V0Indicate initial template three-dimensional people
Face grid vertex set;A indicates the relational matrix of bound term, Bi+1It is the square that the relational matrix end value under NextState is constituted
Battle array;
Third step, fixed Vi+1, in WiOn the basis of the weight coefficient W of NextState is calculated with gradient descent methodi+1, state
Renewal equation such as following formula:
Wherein, f ' (Wi) indicate energy function in WiUnder derivative, be a real number,Indicate real number, the step-length of gradient decline takes
Value is α=0.00000001, due to WiItself is also a dual vector, therefore f function is to the inclined of each weight real variable
Derivative calculations such as following formula:
And single wiChange step β take 0.000001;
4th step updates Vi←Vi+1, Wi←Wi+1, when meeting f ' (Wi)<ε1Or edistance<ε2Condition terminates optimization process, otherwise
It is continued to execute into second step, wherein relative distance error edistanceTo measure three-dimensional face grid vertex collection and mould lack of standardization
Error between plate three-dimensional face grid, ε1Indicate f ' (Wi) upper limit value, ε2Indicate edistanceUpper limit value.
6. the three-dimensional face grid texture method lack of standardization approached as claimed in claim 5 based on template face, feature are existed
In the relative distance error edistanceCalculation method such as following formula:
Wherein, fdistance(Vi) indicate that all vertex project recently to template three-dimensional face grid on three-dimensional face grid lack of standardization
Point square distance and,Indicate on three-dimensional face grid lack of standardization all vertex to the distance of world coordinate system origin
Quadratic sum.
7. the three-dimensional face grid texture method lack of standardization approached as described in claim 1 based on template face, feature are existed
In in step 1), there are three features for the template three-dimensional face grid tool:
1. template three-dimensional face network topology is symmetrical according to face's axis face;
2. the vertex number of template three-dimensional face grid, which less than 20,000, is subject to, is applicable in application and development;
3. the resolution ratio of template texture image is 1024*1024.
8. the three-dimensional face grid texture method lack of standardization approached as claimed in claim 2 based on template face, feature are existed
In: in step 3), there are three features for the specification three-dimensional face grid tool ultimately generated:
1. having five features similar with three-dimensional face grid lack of standardization;
2. having network topology identical with template three-dimensional face grid and vertex number;
3. sharing same template with template three-dimensional face grid parameterizes texture coordinate, texture can be carried out with template texture image
Mapping, mapping relations are still above-mentioned g [f (pi)]=pixeli。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810841353.2A CN108986221B (en) | 2018-07-27 | 2018-07-27 | Non-standard three-dimensional face grid texture method based on template face approximation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810841353.2A CN108986221B (en) | 2018-07-27 | 2018-07-27 | Non-standard three-dimensional face grid texture method based on template face approximation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108986221A true CN108986221A (en) | 2018-12-11 |
CN108986221B CN108986221B (en) | 2022-07-22 |
Family
ID=64551896
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810841353.2A Active CN108986221B (en) | 2018-07-27 | 2018-07-27 | Non-standard three-dimensional face grid texture method based on template face approximation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108986221B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111127668A (en) * | 2019-12-26 | 2020-05-08 | 网易(杭州)网络有限公司 | Role model generation method and device, electronic equipment and storage medium |
CN111340715A (en) * | 2019-09-19 | 2020-06-26 | 杭州海康慧影科技有限公司 | Method and device for weakening grid lines of image and electronic equipment |
WO2020134436A1 (en) * | 2018-12-29 | 2020-07-02 | 华为技术有限公司 | Method for generating animated expression and electronic device |
CN111640055A (en) * | 2020-05-22 | 2020-09-08 | 构范(厦门)信息技术有限公司 | Two-dimensional face picture deformation method and system |
CN112489216A (en) * | 2020-11-27 | 2021-03-12 | 北京百度网讯科技有限公司 | Method, device and equipment for evaluating face reconstruction model and readable storage medium |
CN113487728A (en) * | 2021-07-23 | 2021-10-08 | 中国科学院水生生物研究所 | Fish body model determination method and system |
CN113920282A (en) * | 2021-11-15 | 2022-01-11 | 广州博冠信息科技有限公司 | Image processing method and device, computer readable storage medium, and electronic device |
CN115409933A (en) * | 2022-10-28 | 2022-11-29 | 北京百度网讯科技有限公司 | Multi-style texture mapping generation method and device |
-
2018
- 2018-07-27 CN CN201810841353.2A patent/CN108986221B/en active Active
Non-Patent Citations (1)
Title |
---|
童晶等: "面向影视动画的真实感三维人脸快速建模算法", 《计算机应用》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020134436A1 (en) * | 2018-12-29 | 2020-07-02 | 华为技术有限公司 | Method for generating animated expression and electronic device |
US11410369B2 (en) | 2018-12-29 | 2022-08-09 | Huawei Technologies Co., Ltd. | Method for generating animated expression and electronic device |
CN111340715A (en) * | 2019-09-19 | 2020-06-26 | 杭州海康慧影科技有限公司 | Method and device for weakening grid lines of image and electronic equipment |
CN111340715B (en) * | 2019-09-19 | 2024-02-06 | 杭州海康慧影科技有限公司 | Grid pattern weakening method and device of image and electronic equipment |
CN111127668A (en) * | 2019-12-26 | 2020-05-08 | 网易(杭州)网络有限公司 | Role model generation method and device, electronic equipment and storage medium |
CN111127668B (en) * | 2019-12-26 | 2023-08-22 | 网易(杭州)网络有限公司 | Character model generation method and device, electronic equipment and storage medium |
CN111640055A (en) * | 2020-05-22 | 2020-09-08 | 构范(厦门)信息技术有限公司 | Two-dimensional face picture deformation method and system |
CN111640055B (en) * | 2020-05-22 | 2023-04-11 | 构范(厦门)信息技术有限公司 | Two-dimensional face picture deformation method and system |
CN112489216B (en) * | 2020-11-27 | 2023-07-28 | 北京百度网讯科技有限公司 | Evaluation method, device and equipment of facial reconstruction model and readable storage medium |
CN112489216A (en) * | 2020-11-27 | 2021-03-12 | 北京百度网讯科技有限公司 | Method, device and equipment for evaluating face reconstruction model and readable storage medium |
CN113487728A (en) * | 2021-07-23 | 2021-10-08 | 中国科学院水生生物研究所 | Fish body model determination method and system |
CN113920282A (en) * | 2021-11-15 | 2022-01-11 | 广州博冠信息科技有限公司 | Image processing method and device, computer readable storage medium, and electronic device |
CN115409933B (en) * | 2022-10-28 | 2023-02-03 | 北京百度网讯科技有限公司 | Multi-style texture mapping generation method and device |
CN115409933A (en) * | 2022-10-28 | 2022-11-29 | 北京百度网讯科技有限公司 | Multi-style texture mapping generation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN108986221B (en) | 2022-07-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108986221A (en) | A kind of three-dimensional face grid texture method lack of standardization approached based on template face | |
CN100561523C (en) | A kind of method for re-establishing three-dimensional model gridding | |
CN102982578B (en) | Estimation method for dressed body 3D model in single character image | |
CN101916454B (en) | Method for reconstructing high-resolution human face based on grid deformation and continuous optimization | |
CN104616345B (en) | Octree forest compression based three-dimensional voxel access method | |
US20050140670A1 (en) | Photogrammetric reconstruction of free-form objects with curvilinear structures | |
CN105678683A (en) | Two-dimensional storage method of three-dimensional model | |
CN106846392A (en) | The method and apparatus of three-dimensional modeling | |
CN101303772A (en) | Method for modeling non-linear three-dimensional human face based on single sheet image | |
CN105261062B (en) | A kind of personage's segmentation modeling method | |
CN107657659A (en) | The Manhattan construction method for automatic modeling of scanning three-dimensional point cloud is fitted based on cuboid | |
CN103500467A (en) | Constructive method of image-based three-dimensional model | |
CN110378349A (en) | The mobile terminal Android indoor scene three-dimensional reconstruction and semantic segmentation method | |
CN111028335B (en) | Point cloud data block surface patch reconstruction method based on deep learning | |
CN103530907A (en) | Complicated three-dimensional model drawing method based on images | |
CN112530005B (en) | Three-dimensional model linear structure recognition and automatic restoration method | |
CN110378992A (en) | Towards large scene model web terminal dynamic rendering LOD processing method | |
CN112785692A (en) | Single-view-angle multi-person human body reconstruction method based on depth UV prior | |
CN111127658A (en) | Point cloud reconstruction-based feature-preserving curved surface reconstruction method for triangular mesh curved surface | |
CN109448135A (en) | A method of the threedimensional model repairing based on multi-source data | |
Lai et al. | Surface mosaics | |
CN115861558A (en) | Multistage simplification method for space data model slice | |
Han et al. | 3D reconstruction of dense model based on the sparse frames using RGBD camera | |
Zhou et al. | Progress and review of 3D image feature reconstruction | |
Zhu et al. | Reconstruction of 3D maps for 2D satellite images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |