CN105405163A - Vivid static-state hair modeling method based on multiple direction fields - Google Patents

Abstract

The invention discloses a vivid static-state hair modeling method based on multiple direction fields. The modeling method comprises following steps of 1: based on acquired multi-view hair images, firstly obtaining two-dimensional hair filament segments and hair point cloud models and then generating two-dimensional hair filament segment hair models with directions; 2: according to results of the first step, carrying out direction field solving and keeping regularization of direction fields of structures; and 3: constructing a static-state hair model based on multiple direction fields. The step 3 comprises three stages of generating vivid hair roots, constructing hair filament generating rules and optimizing the three-dimensional hair filament hair models. According to the invention, compared with other hair modeling method based on multi-view images, the modeling method can keep structures of hair in appearance, be approximate to acquisition data, and also allow the inner hair filaments to grow and be distributed in a quite true and reliable manner.

Description

A kind of static scalp electroacupuncture method true to nature based on multi-direction field

Technical field

The invention belongs to virtual reality technology and computer vision field, particularly based on the static scalp electroacupuncture field of multiple image, several hair photos specifically taken by different points of view carry out three-dimensional reconstruction, comprise the field of direction regularization that the three-dimensional hair section Hair model in direction is with in generation, the field of direction solves and retain structure and the static Hair model built based on multi-direction field, be mainly used in the fields such as numbers game, video display animation and advertisement.

Background technology

Scalp electroacupuncture is one of important content of virtual reality and field of Computer Graphics research.Scalp electroacupuncture mainly contains two kinds of methods: one is the scalp electroacupuncture method of physically based deformation, adopting the hair physical model meeting physics law, going out different Hair models by arranging different parametric configurations; Two is the scalp electroacupuncture methods based on image, utilizes the hair data configuration Hair model gathered.

One, the scalp electroacupuncture of physically based deformation

Traditional scalp electroacupuncture constructs according to physical model, comprise (Bertails, F., Audoly, B., Cani, M., Querleux, B., Leroy, F.andL é v ê que, J.2006.Super-helicesforpredictingthedynamicsofnaturalhai r.ACMTransactionsonGraphics (TOG), 25, 3, 1180-1187.) (Selle, A., Lentine, M.andFedkiw, R.2008.Amassspringmodelforhairsimulation.ACMTransactions onGraphics (TOG), 27, 3, 64.) (Derouet-Jourdan, A., Bertails-Descoubes, F., Daviet, G.andThollot, J.2013.Inversedynamichairmodelingwithfrictionalcontact.A CMTransactionsonGraphics (TOG), 32, 6, 159.) (Chai, M., Zheng, C., Xu, W.andZhou, K.2014.AReducedModelforInteractiveHairs.ACMTransactionso nGraphics (TOG), 33, 4, 124.).These methods utilize different physical models to represent Hair model, move afterwards for artificial head shipping.This kind of Method Modeling process is directly perceived not, is difficult to control, and calculated amount is very large, and the model fidelity of structure is not high.The image configuration Hair model that the present invention utilizes multiple views to gather, has the advantages such as modeling speed is fast, model fidelity is high.

Two, based on the scalp electroacupuncture of image

Scalp electroacupuncture method based on image mainly comprises one-view image modeling and multi-view image modeling.(Chai is mainly comprised in the recent period in one-view image modeling area research content, M., Wang, L., Weng, Y., Yu, Y., Guo, B.andZhou, K.2012.Single-viewhairmodelingforportraitmanipulation.AC MTransactionsonGraphics (TOG), 31, 4, 116.) (Chai, M., Wang, L., Weng, Y., Jin, X.andZhou, K.2013.Dynamichairmanipulationinimagesandvideos.ACMTrans actionsonGraphics (TOG), 32, 4, 75.) (HuL, MaC, LuoL, etal.2015.Single-viewhairmodelingusingahairstyledatabase .ACMTransactionsonGraphics (TOG), 34, 4, 125) (ChaiM, LuoL, SunkavalliK, etal.2015.High-qualityhairmodelingfromasingleportraitpho to.ACMTransactionsonGraphics (TOG), 34, 6, 204).The single photo that these methods are taken from single view or video, reconstruction of three-dimensional Hair model.Because single view modeling problem is ill-posed problem in itself, complete real three-dimensional Hair model therefore can not be obtained.The real three-dimensional Hair model of the hair image reconstruction that the present invention adopts multiple views to gather, research method related to this mainly comprises (Paris, S., h.M.andSillion, F.X., 2004.Captureofhairgeometryfrommultipleimages.ACMTransact ionsonGraphics (TOG), 23, 3, 712-719) (Wei, Y., Ofek, E., Quan, L.andShum, H.2005.Modelinghairfrommultipleviews.ACMTransactionsonGr aphics (TOG), 24, 3, 816-820) (Paris, S., Chang, W., Kozhushnyan, O.I., Jarosz, W., Matusik, W., Zwicker, M.andDurand, F.2008.Hairphotobooth:geometricandphotometricacquisition ofrealhairstyles.ACMTransactionsonGraphics (TOG), 27, 3, 30).These class methods are structure grain field first, grows hair afterwards in hair zones from scalp, but owing to not considering the external structure of hair, the hair style of reconstruction can not well retain hair external structure, well can not rebuild the complicated hair styles such as curling, wave.Other class methods are from the external structure of hair, adopt bottom-to-top method to build hair style, comprise (Luo, L., Li, H.andRusinkiewicz, S.2013.Structure-awarehaircapture.ACMTransactionsonGraph ics (TOG), 32,4,76) (Hu, L., Ma, C., Luo, L.andLi, H., 2014.RobustHairCaptureUsingSimulatedExamples.ACMTransact ionsonGraphics (TOG), 33,4,126).These class methods can retain the external structure of hair preferably, but can not ensure that the distribution of root of hair and hair meets truly.

Summary of the invention

The technical problem to be solved in the present invention is: overcome the deficiencies in the prior art, a kind of static scalp electroacupuncture method true to nature based on multi-direction field is provided, the method can keep the external structure of hair, make hair appearance and gather image approximate, root of hair and the distribution of inner hair can be made again more truly, there is higher practical value.

The present invention solves the problems of the technologies described above adopted technical scheme: based on the static scalp electroacupuncture method true to nature of multi-direction field, be divided into 3 stages:

1st stage, generate the three-dimensional hair section Hair model in band direction, this stage first utilization orientation filter operator solve two-dimensional directional field, and solve reliable two-dimentional hair section, then utilize PMVS (Patch-basedMulti-ViewStereo) algorithm to generate the hair cloud data in band direction, finally utilize hair section generating algorithm to generate three-dimensional hair section Hair model;

In 2nd stage, the result based on the 1st stage solves the field of direction and carries out retaining the field of direction regularization of structure, and this stage is divided into 2 steps:

(1) according to the result in the 1st stage, the thermic vibrating screen of belt restraining is utilized to solve internal direction field, inner distance field and the surface structure field of direction respectively;

(2) regularization mapping and diffusion are carried out to the inner field of direction, obtain the regularization field of direction retaining structure;

In 3rd stage, the result based on the 2nd stage builds the static Hair model based on multi-direction field, and this stage is divided into 3 steps:

(1) use and build the method for distribution of shapes, on head model, appointed area generates equally distributed root of hair, and root of hair point direction is determined by the tangent line of this point and normal direction;

(2) use the field of direction retaining structure and Hair model structural constraint, build hair growing strategy, pointwise growth hair;

(3) carry out energy-optimised to the Hair model generated, generate three-dimensional hair Hair model true to nature.

The present invention's advantage is compared with prior art:

(1) the present invention is in the three-dimensional hair section Hair model of generation, employ reliable two-dimentional hair section as restraining growth condition, improve the order of accuarcy of hair pointwise growth, ensure that the three-dimensional hair section Hair model obtained is more similar to the hair image of collection.

(2) the present invention sends out model generation multiple directions field from the three-dimensional hair paragraph header in band direction, and the inner field of direction is used to the field of direction regularization retaining structure, ensure that the inner hair of hair style is evenly distributed, the direction of growth of hair is continuous, can keep hair style external structure simultaneously.

(3) the present invention is at structure based in the static Hair model of multi-direction field, and from equally distributed root of hair, pointwise growth hair, and it is energy-optimised to carry out based on hair to final hair model, makes the Hair model that builds truer

In a word, the present invention can obtain three-dimensional hair section Hair model accurately, can ensure that again the root of hair of final Hair model and the distribution of inner hair meet physical instinct, can also keep the external structure of hair simultaneously, obtain more real three-dimensional Hair model.

Accompanying drawing explanation

Fig. 1 is a kind of static scalp electroacupuncture method flow diagram true to nature based on multi-direction field of the present invention;

Fig. 2 is that the present invention generates three-dimensional hair paragraph header and sends out model result figure;

Fig. 3 is that the present invention generates equally distributed root of hair schematic diagram;

Fig. 4 is the three-dimensional reconstruction design sketch that the present invention is directed to different hair style.

Embodiment

Below in conjunction with accompanying drawing and example, the present invention is described in further detail:

As shown in Figure 1, concrete steps of the present invention following (in dotted line frame, step is prior art):

1, the three-dimensional hair section Hair model in band direction is generated from the multi-view image demarcating rear shooting.

First prior art is utilized to solve its field of direction to the two dimensional image under each viewpoint, recycle two-dimentional hair section tracing algorithm and try to achieve reliable two-dimentional hair section S2D, the field of direction is made up of 32 symmetrical Gabor kernel functions, and direction θ is uniformly distributed between [0, π]:

$K_{\mathrm{\θ}}(u,v)=\exp (-\frac{1}{2}\[\frac{{\tilde{u}}^2}{{\mathrm{\σ}}_u^2}+\frac{{\tilde{v}}^2}{{\mathrm{\σ}}_v^2}\])cos\left(\frac{2\mathrm{\π}\tilde{u}}{\mathrm{\λ}}\right),$ $\tilde{u}=ucos\mathrm{\θ}+vsin\mathrm{\θ},\tilde{v}=-usin\mathrm{\θ}+vcos\mathrm{\θ}$

In above-mentioned kernel function, u, v to be illustrated respectively in single image laterally and vertical on location of pixels, the parameter value that this method uses is σ _u=σ _v=2, λ=3.

Secondly, PMVS (Patch-basedMulti-ViewStereo) algorithm is utilized to generate initial hair cloud data, again in conjunction with the demarcation information of two-dimensional directional field corresponding to each viewpoint and each viewpoint, solve the direction of each point in hair cloud data, form the hair cloud data P in band direction.

Finally, utilize hair section generating algorithm to generate three-dimensional hair section Hair model, the three-dimensional hair paragraph header of generation sends out model result as shown in Figure 2, and the method comprises following sub-step:

1.1 generate Seed Points { SP _i}

Seed Points { the SP that confidence level is greater than the some combination producing three-dimensional hair section Hair model of a certain threshold value (as 0.5) is chosen from the hair point cloud P that PMVS algorithm generates _i, and by Seed Points by the descending sequence of confidence level.

1.2 from a Seed Points, tracks out one section of three-dimensional hair section S

Three-dimensional hair section S is by point { p a series of in three dimensions ₁, p ₂..., p _nbe formed by connecting, by following the trail of the point in the three-dimensional hair section of operation generation one as follows: 1) for a Seed Points SP _i, first try to achieve the natural on-off cycles of hair growth direction t (SP at this Seed Points place _i)=± O (SP _i), O (SP _i) be SP in hair point cloud P _ithe three-dimensional field at place; 2) for new some p of one step growth every in hair section S _i, when growth step-length is δ (δ=1.0mm), try to achieve the position p of interim new point _{i_temp}=p _i+ t (SP _i) δ; 3) p is tried to achieve by Moving Least MLS function _{i_temp}at the new some p that a cloud P is corresponding on the surface _{i_MLS}=MLS (p _{i_temp}, P); 4) by two-dimentional hair section correction function to p _{i_MLS}carry out the correction of reliable two-dimentional hair section, obtain revised some p _{i_2DCorr}, correction function is defined as p _{i_2DCorr}=Tra2DCorr (p _{i_MLS}, V _j, S2D), wherein, V _jfor the multi-view image of all demarcation, S2D is the reliable two-dimentional hair section of all generations, and its processing procedure is: for input point p _{i_MLS}, first projected to all V _jin, if this point can be found in reliable two-dimentional hair section S2D, be then reliable growing point, p _{i_2DCorr}=p _{i_MLS}, otherwise by minimization of energy function revise, V _jrepresent a jth visual point image, image V _jprojection matrix be pass through formula try to achieve, wherein (x ₁, x ₂, x ₃) be p _{i_MLS}corresponding three dimensions point, it is distance two-dimensional projection point on S2D closest approach in threshold value 0.1mm, final revised point is 5) to p _{i_2DCorr}moving Least MLS function is used to obtain final some p _{i_final}=MLS (p _{i_2DCorr}, P), if meet following two conditions, then p _{i_final}stop growth: 1) | O (p _{i_final}) O (p _i) | < 0.9, some cloud P mid point p _{i_final}with a p _idirection O (the p at place _{i_final}) and O (p _i) incompatible, 2) represent some cloud P mid point p _{i_final}the quantity that place around puts; Otherwise will p be put _{i_final}join in three-dimensional hair section S, the new point of continued growth, until meet end condition; 6) the smoothing operation of three-dimensional hair section S will generated, and the S after level and smooth is added in three-dimensional hair section Hair model S3D, simultaneously by Seed Points SP _ifrom { SP _iremove.

1.3 orientation consistency calculate

Repeat step 1.2, until { SP _iin till all Seed Points are all removed.Finally calculate the three-dimensional hair section Hair model S3D travel direction consistance generated, make the direction of every root hair S be all along gravity direction growth, edge is away from the growth of scalp direction simultaneously, to every root hair definition hair energy as shown in formula (1):

E(S)＝H(S ₀)-H(S _l)+Dis(S ₀)-Dis(S _l)(1)

In formula (1), S ₀and S _lcorresponding length is the upper point of the 1st position of hair S and the point of l position of l respectively, and H is a depth function, and for the p of on hair S, its degree of depth energy is H (p)=pV _down, V _downbe from root of hair to the reference direction sending out the tip on hair, Dis (p) represents that on hair S, some p is to the distance of nearest root of hair point.When hair ENERGY E (S) is less than or equal to 0, represent that on hair S, point is from root of hair to sending out tip arrangement, when hair ENERGY E (S) is greater than 0, needing the point on hair S to carry out reverse permutatation, guaranteeing that the direction of all hairs is all arranged to sending out the tip by root of hair.

2, multiple directions field is solved

Obtain three-dimensional regular grid by hair cloud data, more three-dimensional hair paragraph header is sent out Model Mapping on regularization grid, obtain internal direction field by thermic vibrating screen, solve inner distance field and the surface structure field of direction, step is as follows simultaneously:

The first step, solves hair zones Bounding Box by hair cloud data P, and it is discretely turned to three-dimensional regular grid, and grid resolution is set to 2mm;

Second step, three-dimensional hair section Hair model S3D is mapped on regularization grid, build internal direction field V, the direction at the net point place of three-dimensional hair process is determined do not have the direction of the net point of three-dimensional hair process to pass through thermic vibrating screen by the tangential direction of this point on hair determine, the net point of known direction as boundary constraint, V=vv ^t, v is the direction at net point place;

3rd step, from three-dimensional hair section Hair model S3D, generate the hair belt Ribbon retaining hair partial structurtes according to existing method, after travel direction consistent problem solving, the Ribbon in band direction is mapped on regularization grid, builds surface structure field of direction S _dir, the direction at the net point place having hair band Ribbon to pass through is determined by the tangential direction of this point on hair band, and all the other unknown point are tried to achieve by solving thermic vibrating screen;

4th step, inner distance field I is solved in three-dimensional regular grid inside, be 0 at the Grid point Value of scalp surface process, the Grid point Value of three-dimensional hair section Hair model S3D process is 1, the value that net boundary is put is that the distance of nearest hair point on this net point to three-dimensional hair section Hair model S3D adds 1, and the direction of all the other points is tried to achieve by thermic vibrating screen.

3, the field of direction regularization of structure is retained

The field of direction regularization retaining structure mainly carries out regularization mapping to inner field of direction V, and making in inner distance field I value is that 1 place's external structure can retain, and the inside hair being less than 1 place in inner distance field I intermediate value is level and smooth, and step is as follows:

The first step, is first converted into structure tensor V=vv to the direction v at each net point place of hair internal direction field V ^t, then the eigenvalue of maximum characteristic of correspondence vector of symmetric matrix V is still the direction v at net point place;

Second step, carries out convolution by the V obtained in the first step by sphere three-dimensional Gaussian core K (σ) and obtains λ ₁>=λ ₂>=λ ₃v _σthe eigenwert of corresponding descending arrangement, its average is definition rule is mapped as $m=M\left\{V_{\mathrm{\&sigma;}}\right\}=\frac{\sqrt{3}}{\sqrt{2}}\frac{\sqrt{{({\mathrm{\&lambda;}}_1-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2+{({\mathrm{\&lambda;}}_2-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2+{({\mathrm{\&lambda;}}_3-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2}}{\sqrt{{\mathrm{\&lambda;}}_1^2+{\mathrm{\&lambda;}}_2^2+{\mathrm{\&lambda;}}_3^2}},$ M can embody desired structure;

3rd step, definition rule diffusion matrix is $D=\left[\begin{array}{ccc}{u}_1& u_2& u_3\end{array}\right]\left[\begin{array}{ccc}{\mathrm{\&mu;}}_1& 0& 0\\ 0& {\mathrm{\&mu;}}_2& 0\\ 0& 0& {\mathrm{\&mu;}}_3\end{array}\right]\left[\begin{array}{c}{u}_1^T\\ u_2^T\\ u_3^T\end{array}\right],$ μ _i(i=1,2,3) determine along u _ithe smoothness that (i=1,2,3) spread;

4th step, according to the regularization diffusion matrix D that the 3rd step obtains, each component value upgraded in structure tensor V is the proper vector v of the corresponding eigenvalue of maximum of V is the final direction at each net point place on regularization grid.

4, root of hair true to nature is generated

Generate root of hair true to nature and mainly comprise following three steps: first, manually selected hair zones calculate the total area in area corresponding to each tri patch and natural on-off cycles of hair growth region; Secondly on each tri patch, equally distributed root of hair is generated by area ratio; Finally give hairy root growth direction to the root of hair generated, the root of hair that final acquisition is true to nature, generate equally distributed root of hair as shown in Figure 3, concrete steps are as follows:

The first step, manually chooses natural on-off cycles of hair growth region in the head model that tri patch represents, calculates the area S that each tri patch is corresponding _i(i=1,2 ..., N _{total_patch}) and the total area of whole tri patchs corresponding to natural on-off cycles of hair growth region $S_{total\_patch}=\underset{i=1}{\overset{N_{total\_patch}}{\&Sigma;}}{S}_i;$

Second step, the hair total quantity that whole Hair model is corresponding is N _{total_hair}, then the number of hairs that each tri patch is corresponding is N _{i_hair}=N _{total_hair}* (S _i/ S _{total_patch}) (i=1,2 ..., N _{total_patch}), be (Vertex on summit ₀, Vertex ₁, Vertex ₂) i-th tri patch on stochastic generation N _{i_hair}individual root of hair point, the position of each root of hair point by $P_{hair\_root}=(1-\sqrt{{\mathrm{rand}}_1}){\mathrm{Vertex}}_0+\sqrt{{\mathrm{rand}}_1}(1-{\mathrm{rand}}_2){\mathrm{Vertex}}_1+\sqrt{{\mathrm{rand}}_1}{\mathrm{rand}}_2{\mathrm{Vertex}}_2$ Determine, rand ₁and rand ₂between 0 to 1, and be set to summit Vertex ₀to limit, opposite Vertex ₁vertex ₂number percent, rand ₂represent along limit Vertex ₁vertex ₂number percent, use rand ₁square root can obtain corresponding to the uniform stochastic sampling root of hair point of tri patch surface area;

3rd step, try to achieve the normal direction at each root of hair point place and tangential respectively, setting root of hair inclination factor inclineFactor, tries to achieve the natural on-off cycles of hair growth direction at each root of hair point place, make its normal direction and tangential between, finally generate root of hair true to nature.

5, hair growing strategy is built

The hair band Ribbon retaining hair partial structurtes can be converted into hair bundle Wisp according to existing method, the disconnected hair section NAS to scalp is obtained from hair bundle Wisp, NAS can retain the partial structurtes of hair, use the field of direction and Hair model structural constraint that retain structure, build hair growing strategy, the step that pointwise grows single hair is as follows:

The first step, every root hair section S={p ₁, p ₂..., p _nbe formed by connecting by continuous print point a series of in three dimensions, p ₁represent root of hair point, for the some p of each new growth _i, its reposition is determined by formula (2),

$p_i=p_{i-1}+\mathrm{\&delta;}\left({\mathrm{\&alpha;}}_{1}V\right(p_{i-1})+{\mathrm{\&alpha;}}_2\&dtri;(I\left(p_{i-1}\right))+{\mathrm{\&alpha;}}_3{O}_{grow}(p_{i-1})+{\mathrm{\&alpha;}}_4{S}_{dir}(p_{i-1}\left)\right)---\left(2\right)$

Wherein, p _i-1p on hair section S _ithe position of previous point, V (p _i-1) and I (p _i-1) be illustrated respectively in a p _i-1the internal direction field value at place and inner distance field value, represent at a p _i-1the inner distance field gradient value at place, O _grow(p _i-1) represent position p on hair section S _i-1the direction of growth at place, the direction of growth of upper first point of S is determined by root of hair direction, and the direction of growth at S all the other some places upper is by (p _i-1-p _i-2) decide, S _dir(p _i-1) be range points p in hair surface _i-1locate the value at nearest surface structure field of direction place, α ₁, α ₂, α ₃, α ₄be the weighted value of dynamic change, correspond respectively to each affined field of direction;

Second step, newly puts p to solve _i, first calculation level p _i-1the distance d of closest approach to hair band Ribbon _sur, according to d _survalue, α ₁, α ₂, α ₄value dynamic change, if d _sur<T _s(T _sbe threshold value, be set to 0.1mm), then α is set ₁=0.1, α ₂=0.1* α ₁, α ₄=100-d _sur, otherwise, if d _sur>=T _s, then α is set ₁=100*d _sur, α ₂=0.1* α ₁, α ₄=0, α ₃constant, α ₃=10;

3rd step, repeated growth newly puts p _iif, the some p of new growth _iposition exceedes hair borderline region, that is: I (p _i) >I _bound, (get threshold value I _bound=1.1), or growth Strand S length reach maximum length, then stop growing;

4th step, for retaining the CONSTRUCTED SPECIFICATION of outside hair, is newly being put p _iafterwards, from NAS, distance p is searched _inearest some q, if meet following two conditions (i) || p _i-q||< △, (ii) | O _grow(p _i) O _grow(q) | >T _o, then q point in the hair section of q place and later all points thereof are all joined in the hair section S of new growth, △ and T _obe threshold value, be set to △=0.2mm respectively according to practical experience, T _o=0.7.

6, three-dimensional hair Hair model is optimized

For generating three-dimensional hair Hair model true to nature, build energy term to each the root hair S in the Hair model generated, use least square method to solve energy optimization problem, try to achieve the reposition that hair is put, concrete steps are:

The first step, builds regularization energy term E _reg, e _regguarantee newly-generated some p _ias far as possible near initial position

Second step, builds the consistent energy term E in direction _ort, e _ortguarantee that the growth of hair is consistent with internal direction field V as far as possible, δ is growth step-length, the same formula of its value (2);

3rd step, builds neighbours energy term E _nei, e _neimake hair S consistent with the hair around it as far as possible, it is distance at R _p(R _p=1mm) distance on each hairline in all hairlines in scope the set of nearest point, o _growq () represents the direction of growth at q point place on hair, w _totalall weight w _qsummation, get σ _d=2.0;

4th step, builds smoothed energy item E _smth, e _smthmake hair S on the whole can be level and smooth;

5th step, total energy term E _totalobtained by four weighted accumulations, E _total=α _rege _reg+ α _orte _ort+ α _neie _nei+ α _smthe _smth, α _reg, α _ort, α _nei, α _smthbe the weights corresponding respectively to each energy term, be set to α respectively according to the percentage contribution of gross energy _reg=0.1, α _ort=2, α _nei=5, α _smth=10, finally use least square method to solve energy optimization problem, can obtain the position of all new points on hair S, the three-dimensional reconstruction effect for different hair style finally obtained as shown in Figure 4.

The content be not elaborated in instructions of the present invention belongs to the known prior art of professional and technical personnel in the field.

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (7)

1., based on a static scalp electroacupuncture method true to nature for multi-direction field, it is characterized in that: described modeling method is divided into 3 stages, specific as follows:

1st stage, generate the three-dimensional hair section Hair model in band direction, this stage first utilization orientation filter operator solve two-dimensional directional field, and solve reliable two-dimentional hair section, then utilize PMVS (Patch-basedMulti-ViewStereo) algorithm to generate the hair cloud data in band direction, finally utilize hair section generating algorithm to generate the three-dimensional hair section Hair model in band direction;

In 2nd stage, the result based on the 1st stage solves the field of direction and carries out retaining the regularization of structure, and this stage is divided into 2 steps:

(1) according to the result in the 1st stage, the thermic vibrating screen of belt restraining is utilized to solve internal direction field, inner distance field and the surface structure field of direction respectively;

(2) regularization mapping and diffusion are carried out to the inner field of direction, obtain the regularization field of direction retaining structure;

In 3rd stage, the result based on the 2nd stage builds the static Hair model based on multi-direction field, and this stage is divided into 3 steps:

(1) use and build the method for distribution of shapes, on head model, appointed area generates equally distributed root of hair, and root of hair point direction is determined by the tangent line of this point and normal direction;

(2) use the field of direction retaining structure and Hair model structural constraint, build hair growing strategy, pointwise growth hair;

(3) carry out energy-optimised to the Hair model generated, generate three-dimensional hair Hair model true to nature.

2. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, is characterized in that: the method utilizing hair section generating algorithm to generate three-dimensional hair section Hair model in described 1st stage comprises following sub-step:

Step (A1), from the hair point cloud P that PMVS algorithm generates, choose the Seed Points { SP that confidence level is greater than the some combination producing three-dimensional hair section Hair model of a certain threshold value _i, and by Seed Points by the descending sequence of confidence level;

Step (A2), from a Seed Points SP _iset out, tracking out one section of three-dimensional hair section S, S is by point { p a series of in three dimensions ₁, p ₂..., p _nbe formed by connecting, by following the trail of the point in the three-dimensional hair section of operation generation one as follows:

Step (A2.1), for a Seed Points SP _i, first try to achieve the natural on-off cycles of hair growth direction t (SP at this Seed Points place _i)=± O (SP _i), O (SP _i) be SP in the hair point cloud P in band direction _ithe three-dimensional field at place;

Step (A2.2), new some p for one step growth every in hair section S _i, when growth step-length is δ, try to achieve the position p of interim new point _{i_temp}=p _i+ t (SP _i) δ;

Step (A2.3), try to achieve p by Moving Least MLS function _{i_temp}at the new some p that a cloud P is corresponding on the surface _{i_MLS}=MLS (p _{i_temp}, P);

Step (A2.4), by two-dimentional hair section correction function to p _{i_MLS}carry out the correction of reliable two-dimentional hair section, obtain revised some p _{i_2DCorr}, correction function is defined as p _{i_2DCorr}=Tra2DCorr (p _{i_MLS}, V _j, S2D), wherein, V _jfor the multi-view image of all demarcation, S2D is the reliable two-dimentional hair section of all generations, and its processing procedure is: for input point p _{i_MLS}, first projected to all V _jin, if this point can be found in reliable two-dimentional hair section S2D, be then reliable growing point, p _{i_2DCorr}=p _{i_MLS}, otherwise by minimization of energy function revise, V _jrepresent a jth visual point image, image V _jprojection matrix be pass through formula try to achieve, wherein (x ₁, x ₂, x ₃) be p _{i_MLS}corresponding three dimensions point, it is distance two-dimensional projection point on S2D closest approach in threshold value 0.1mm, final revised point is $p_{i\_2DCorr}=M_{V_j}^{-1}{(u_{V_j},v_{V_j},d_{V_j})}^T;$

Step (A2.5), to p _{i_2DCorr}least square method MLS function is used to obtain final some p _{i_final}=MLS (p _{i_2DCorr}, P), if meet following two conditions, then p _{i_final}stop growth: 1) | O (p _{i_final}) O (p _i) | <0.9, some cloud P mid point p _{i_final}with a p _idirection O (the p at place _{i_final}) and O (p _i) incompatible, 2) represent some cloud P mid point p _{i_final}the quantity that place around puts; Otherwise will p be put _{i_final}join in three-dimensional hair section S, the new point of continued growth, until meet end condition;

Step (A3), the smoothing operation of three-dimensional hair section S that will generate, and the S after level and smooth is added in three-dimensional hair section Hair model S3D, simultaneously by Seed Points SP _ifrom { SP _iremove;

Step (A4), repetition step (A2) (A3), until { SP _iin till all Seed Points are all removed;

Step (A5), the three-dimensional hair section Hair model S3D travel direction consistance generated being calculated, making the direction of every root hair S be all along gravity direction growth, simultaneously along growing away from scalp direction.

3. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, is characterized in that: the method solving multiple directions field in the step (1) in described 2nd stage comprises the following steps:

Step (B1), solve hair zones Bounding Box by hair cloud data P, and it is discretely turned to three-dimensional regular grid, grid resolution is set to 2mm;

Step (B2), three-dimensional hair section Hair model S3D is mapped on regularization grid, build internal direction field V, the direction at the net point place of three-dimensional hair process is determined do not have the direction of the net point of three-dimensional hair process to pass through thermic vibrating screen by the tangential direction of this point on hair determine, the net point of known direction as boundary constraint, V=vv ^t, v is the direction at net point place;

Step (B3), obtain by three-dimensional hair section Hair model S3D the hair band Ribbon retaining hair partial structurtes according to local similarity, after travel direction consistent problem solving, the Ribbon in band direction is mapped on regularization grid, builds surface structure field of direction S _dir, same to step (B2), the direction at the net point place having hair band to pass through is determined by the tangential direction of this point on hair band, and all the other unknown point are then tried to achieve by solving thermic vibrating screen;

Step (B4), solve inner distance field I in three-dimensional regular grid inside, the Grid point Value of scalp surface process is 0, the Grid point Value of three-dimensional hair section Hair model S3D process is 1, the value that net boundary is put is that the distance of nearest hair point on this net point to three-dimensional hair section Hair model S3D adds 1, and the direction of all the other points is tried to achieve by thermic vibrating screen.

4. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, is characterized in that: the field of direction rule method retaining structure in the step (2) in described 2nd stage comprises the following steps:

Step (C1), by the direction v at each net point place of hair internal direction field V that obtains in step B2 by structure tensor product V=vv ^ttry to achieve, (V, v, V represent different implications to symmetric matrix V respectively, and V represents vector field, such as the field of direction, are all to represent by capitalization black matrix italic, and general vector small letter black matrix represents, such as direction v; Capitalization black matrix is used for representing matrix, such as V and V _σ) eigenvalue of maximum characteristic of correspondence vector be still the direction v at net point place;

Step (C2), the V obtained is carried out convolution by sphere three-dimensional Gaussian core K (σ) obtain in step C1 λ ₁>=λ ₂>=λ ₃v _σthe eigenwert of corresponding descending arrangement, its average is definition rule is mapped as $m=M\left\{V_{\mathrm{\&sigma;}}\right\}=\frac{\sqrt{3}}{\sqrt{2}}\frac{\sqrt{{({\mathrm{\&lambda;}}_1-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2+{({\mathrm{\&lambda;}}_2-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2+{({\mathrm{\&lambda;}}_3-\stackrel{\&OverBar;}{\mathrm{\&lambda;}})}^2}}{\sqrt{{\mathrm{\&lambda;}}_1^2+{\mathrm{\&lambda;}}_2^2+{\mathrm{\&lambda;}}_3^2}},$ M can embody desired structure;

Step (C3), definition rule diffusion matrix are $D=\&lsqb;\begin{array}{ccc}{u}_1& u_2& u_3\end{array}\&rsqb;\left[\begin{array}{ccc}{\mathrm{\&mu;}}_1& 0& 0\\ 0& {\mathrm{\&mu;}}_2& 0\\ 0& 0& {\mathrm{\&mu;}}_3\end{array}\right]\left[\begin{array}{c}{u}_1^T\\ u_2^T\\ u_3^T\end{array}\right],$ μ _i(i=1,2,3) determine along u _ithe smoothness that (i=1,2,3) spread;

Step (C4), the regularization diffusion matrix D obtained according to step (C3), each component value upgraded in structure tensor V is wherein the proper vector v of the corresponding eigenvalue of maximum of i=1...9, V is the final direction at each net point place on regularization grid.

5. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, it is characterized in that: in the step (1) in described 3rd stage, use the method building distribution of shapes to generate root of hair true to nature, be specially: first, manually selected hair zones calculate the total area in area corresponding to each tri patch and natural on-off cycles of hair growth region; Secondly on each tri patch, equally distributed root of hair is generated by area ratio; Finally give hairy root growth direction to the root of hair generated, the root of hair that final acquisition is true to nature.

6. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, it is characterized in that: in the step (2) in described 3rd stage, use the field of direction retaining structure and Hair model structural constraint, build hair growing strategy, pointwise growth hair comprises the following steps:

Step (E1), the hair band Ribbon of reservation hair partial structurtes step (B3) obtained are converted into hair bundle Wisp, hair bundle Wisp is converted into the disconnected hair section NAS to scalp, NAS can retain the partial structurtes of hair again;

Step (E2), every root hair section S={p ₁, p ₂..., p _nbe formed by connecting by continuous print point a series of in three dimensions, p ₁represent root of hair point, for the some p of each new growth _i, its reposition is by formula $p_i=p_{i-1}+\mathrm{\&delta;}\left({\mathrm{\&alpha;}}_{1}V\right(p_{i-1})+{\mathrm{\&alpha;}}_2\&dtri;(I\left(p_{i-1}\right))+{\mathrm{\&alpha;}}_3{O}_{grow}(p_{i-1})+{\mathrm{\&alpha;}}_4{S}_{dir}(p_{i-1}\left)\right)$ Determine, wherein, p _i-1p on hair section S _ithe position of previous point, V (p _i-1) and I (p _i-1) be illustrated respectively in a p _i-1the internal direction field value at place and inner distance field value, represent at a p _i-1the distance field Grad at place, O _grow(p _i-1) represent position p on hair section S _i-1the direction of growth at place, the direction of growth of upper first point of S is determined by root of hair direction, and the direction of growth at S all the other some places upper is by (p _i-1-p _i-2) decide, S _dir(p _i-1) be range points p in hair surface _i-1locate the value at nearest surface structure field of direction place, α ₁, α ₂, α ₃, α ₄be the weight corresponding respectively to each direction that is tied, the new point of repeated growth, until S reaches maximum length or newly puts p _itill exceeding hair zones;

Step (E3), for retaining the CONSTRUCTED SPECIFICATION of outside hair, newly put p _iafterwards, from NAS, distance p is searched _inearest some q, if meet following two conditions (i) || p _i-q||< △ (ii) | O _grow(p _i) O _grow(q) | >T _o, then q point in the hair section of q place and later all points thereof are all joined in the hair section S of new growth, △ and T _obe threshold value, be set to △=0.2mm respectively, T _o=0.7;

Step (E4), repetition step (E2) (E3), until all reach end condition from all hairs of hairy root growth.

7. a kind of static scalp electroacupuncture method true to nature based on multi-direction field according to claim 1, it is characterized in that: for generating three-dimensional hair Hair model true to nature in the step (3) in described 3rd stage, below energy-optimised step is carried out to Hair model: to each the root hair in the Hair model generated, build regularization energy term E respectively _reg, the consistent energy term E in direction _ort, neighbours energy term E _neiwith level and smooth energy term E _smth, total energy term E _totalobtained by four weighted accumulations, E _total=α _rege _reg+ α _orte _ort+ α _neie _nei+ α _smthe _smth, α _reg, α _ort, α _nei, α _smthbe the weights corresponding respectively to each energy term, be set to α respectively according to the percentage contribution of gross energy _reg=0.1, α _ort=2, α _nei=5, α _smth=10, finally use least square method to solve this energy optimization problem.