CN101882326A - Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people - Google Patents

Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people Download PDF

Info

Publication number
CN101882326A
CN101882326A CN2010101806286A CN201010180628A CN101882326A CN 101882326 A CN101882326 A CN 101882326A CN 2010101806286 A CN2010101806286 A CN 2010101806286A CN 201010180628 A CN201010180628 A CN 201010180628A CN 101882326 A CN101882326 A CN 101882326A
Authority
CN
China
Prior art keywords
skull
face
soft tissue
shape
dimensional
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.)
Pending
Application number
CN2010101806286A
Other languages
Chinese (zh)
Inventor
袁中标
裴玉茹
刘超
查红彬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
GUANGZHOU CITY FORENSIC SCIENCE TECHNOLOGY INSTITUTE
Original Assignee
GUANGZHOU CITY FORENSIC SCIENCE TECHNOLOGY INSTITUTE
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by GUANGZHOU CITY FORENSIC SCIENCE TECHNOLOGY INSTITUTE filed Critical GUANGZHOU CITY FORENSIC SCIENCE TECHNOLOGY INSTITUTE
Priority to CN2010101806286A priority Critical patent/CN101882326A/en
Publication of CN101882326A publication Critical patent/CN101882326A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention discloses a three-dimensional craniofacial reconstruction method based on the overall facial structure shape data of Chinese people, comprising the following steps of: (1) acquiring a thickness distribution model of a facial soft tissue based on the statistic analysis of a large amount of human head CT (Computed Tomography) data; (2) unfolding the soft tissue layer and the skull surface layer of a human head through a cylinder, and projecting the soft tissue layer and the skull surface layer to a two-dimensional flat surface, representing the soft tissue shape and the skull shape by using a two-dimensional depth map, and training a radial basis function network to realize the change between the skull to be reconstructed and the common soft tissue thickness distribution form; (3) constructing a shape subspace of local organs of the reconstructed face based on the craniofacial facial structure shape classification of human species, and learning the mapping between the craniofacial local shape and the local shape of the reconstructed face; (4) correcting the reconstructed face model by combining the integral soft tissue distribution and the local characteristic shape deformation, i.e. completing three-dimensional craniofacial reconstruction and correctness of the input skull by adding the two-dimensional depth maps of the soft tissue shape and the shape of the skull to be reconstructed; and (5) synthesizing a complete texture graph of the face through facial texture mapping by using orthogonal pictures, and rendering the skin color and the hair style of the reconstructed human picture to enhance the reality sense of the human picture. By applying the method, the problems of missing reconstruction details and lack of individual characteristics are solved, the manufacturing cost of other anthropology researches and reconstruction technologies is saved, and the working efficiency is improved. The invention has high automation degree and simple operation.

Description

Three-dimensional cranium face restored method based on Chinese's full-face structure graphic data
Technical field
The present invention relates to the three-dimensional cranium face reconstructing method of forensic anthropology and field of Computer Graphics, particularly a kind of three-dimensional cranium face restored method based on Chinese's full-face structure graphic data.
Background technology
For a long time, in the criminal investigation application facet, it is to solve to cut the main method of searching a difficult problem with the corpse source of decompose to bony skeleton to pieces that the cranium face restores, and is the important criminal technique that breaks through difficult case.Traditional Ge Laximofu sculpture restoring method, soft tissue thickness data with the method joint head face of artificial sculpture are restored, the soft tissue thickness data mean value that present domestic recovery work uses document in the past to deliver substantially carries out recovery work, do not consider cranium face structural relation, recovery effect is generally relatively poor.
Innovation development Computerized three-dimensional cranium face restoring method on conventional art, for current criminal investigation work, this achievement in research not only can penetration and promotion, and to save cost and increasing work efficiency than Traditional Sculpture Art restoring method, the emulation that uses a computer is restored can not bring any damage to the skull with important material evidence meaning, have important criminal investigation meaning.In addition, the achievement of this research is also significant in other field, as: in the archaeology field, available three-dimensional computer cranium face restoring method recover the primitive man and ancient times the personage the morphosis looks; In surgery medical science, significant for the planning of facial plasty operation estimation.
The cranium face reconstructing system of existing Computer Simulation generally all be based on a few features point with and corresponding tissue thickness's constraint rebuild mask.Archer has proposed the cranium face reconstructing system based on hierarchical B-splines, and interpolation people's face signature defines and obtains more surface indicia point, sets up the control mesh of spline surface.By increasing the level of control net, Deformation control is come the shape of meticulous each organ of adjustment in the scope of a part.This method of Multi-B Spline curve is used for the local detail shape control of curve at first, has expanded on the curved surface afterwards.The process that parametric surface fits is complicated, needs to prepare meticulously the head template model simultaneously.People such as Vanezis set up the face characteristic 3 d model library by 3-D scanning, from model bank, choose suitable template three-dimensional model for new input skull, make the thickness of feature locations satisfy the statistics of tissue thickness by distortion, on the three-dimensional model that generates, cover photo, the cranium face soverlay technique in the whole process simulation forensic anthropology.People such as Wilhelms propose moulding and the animation system based on the dissect physiology structure, the skeletal muscle epidermis uses spheroid, reduced representation such as cylinder or ellipsoid mode, the muscle that uses a small amount of parameter just can portray between bone insertion point and epidermis connects and the relevant motion of control, contour surface is extracted in the bottom layer tissue voxelization obtain the epidermis polygonal grid model, parameterized bottom assembly can be multiplexing between different models.But, people's face has the part of complex physiologic structure, some positions wherein can not get embodying on bone, it is more coarse usually directly to fit the reconstruction result that universal model obtains by a few unique point, the individual character that lacks people's face, do not reach the condition of Identification of Images identity, lose the meaning of investigation the dead identity.
For example, in disclosed Chinese invention patent application CN 101339670A on January 7th, 2009, a kind of computer assisted three-dimensional cranium face restored method is disclosed, promptly at first set 36 anthropology measurement points of women's head-ornaments portion, then by a large amount of live body head parts of x-ray scanning, add up the soft tissue thickness data of above measuring point, by the skull model of spatial digitizer scanning parked, the soft tissue thickness statistics that covers measuring point is thereon come reconstruct three dimensional face model at last.
Summary of the invention
At above-mentioned defective of the prior art, the present invention is directed to a large amount of live body cranium face CT data and carry out the research of cranium face structural relation, find and obtain the rule of the soft tissue feature formation of its full-face, then according to the cranium face reconstructing method of the craniofacial relation rule design personalized of obtaining ethnic group, thus the three-dimensional cranium face restored method of having set up a kind of precise and high efficiency, can having applied.
In order to achieve the above object, the technical solution adopted in the present invention: a kind of three-dimensional cranium face restored method based on Chinese's full-face structure graphic data may further comprise the steps:
(1) based on a large amount of number of people CT data statistic analysis, obtains facial soft tissue thickness distributed model.Extract the middle soft tissue layer of holocephatic, launch it is projected two dimensional surface, use the depth map of two dimension to represent this organized layer by cylinder;
(2) utilize the automatic extractive technique of feature templates to obtain feature mark in the skull depth map;
(3) the cranium face structure shape based on ethnic group is classified, structure restores the shape subspace of people's face local organs, and the mapping between study skull local form and the recovery people face local form, use radial primary function network to estimate local feature from the bone local measurement of bottom;
(4) for the skull of parked cranium face, with its parametrization to two dimensional surface, analyze its morphological feature, seek approximate tissue thickness's distributed model, at the radial primary function network realization input skull form of particular anatomy character shape training and the conversion between the general soft tissue thickness distribution plan, tissue thickness's distribution plan after the conversion is added on the input skull, and the three-dimensional cranium face that uses two-dimentional depth map addition to finish the input bone is rebuild;
(5) by face texture mapping, use the synthetic facial complete texture maps of orthogonal pictures, and the colour of skin, the hair style of restoring portrait play up, strengthen the sense of reality of portrait.
Above-mentioned steps (1) further may further comprise the steps:
(a) from the CT view data, generate skull and faceform;
(b) with the skull of correspondence and faceform's parametrization to two dimensional surface, use specified resolution to carry out double sampling, people's face and the skull depth map with respect to the sampling central shaft is obtained in sampling;
(c) skull by correspondence and the figure of facial tissue subtract each other and obtain soft tissue chart;
(d) use radial basis function that all organization charts in the same class are carried out registration according to its feature, obtain general soft tissue thickness definition by getting mean value computation.
Compared with prior art, the invention has the advantages that: (1) uses two-dimentional depth map to preserve head tissue information, by obtaining tissue thickness's information of input head skeletal framework lattice arbitrary node position, there is not special requirement for input head skeletal framework lattice topological structure and grid precision to this figure sampling; (2),, just can obtain and import the cranium surface model of bone correspondence by simple superposition with its general skull figure and organization chart use radial primary function network registration with same precision by to input skull sampling parameterization; (3), estimate personalized facial shape based on each structure shape characteristic distribution of skull.
The present invention only need can provide the report of the dead's facial feature in 1 hour and restore picture and give the use of criminal investigation department, and similarity reaches more than 85%, meets the individual identification condition.The cost of other anthropological studies and recovery technique making has been saved in application of the present invention, and has improved work efficiency.Have the automaticity height, characteristics such as easy and simple to handle, non-cranium face restores the professional person also can be operated; Also have prospect in other sector application, as in surgery medical science, facial plasty operation estimation planning etc.
Description of drawings
Fig. 1 is that the individualized feature of the three-dimensional people of recovery face of the present invention obtains process flow diagram;
Fig. 2 is the process flow diagram of cranium face restored method of the present invention.
Embodiment
Below in conjunction with accompanying drawing concrete structure of the present invention is further described.
As shown in Figure 1, 2, the three-dimensional cranium face restored method based on the CT data of the present invention may further comprise the steps: 1, extract facial soft tissue thickness and distribute, and the structure statistical model
The head model that is used to calculate the soft tissue layer thickness information is from the CT image, and soft tissue chart's leaching process step is as follows:
(a) from the CT view data, generate skull and faceform, a large amount of disclosed kits are arranged for the CT Flame Image Process, VTK for example, and based on the VolView software of VTK, be used for rebuilding mask from the CT image;
(b) with the skull of correspondence and faceform's parametrization to two dimensional surface, use specified resolution to carry out double sampling, people's face and the skull depth map with respect to the sampling central shaft is obtained in sampling;
(c) skull by correspondence and the figure of facial tissue subtract each other and obtain soft tissue chart because skull has adopted identical sample mode with the faceform, can be directly by the depth value of respective pixel subtract each other obtain in the middle of the thickness of soft tissue layer define;
(d) use radial basis function that all organization charts in the same class are carried out registration according to its feature, obtain general soft tissue thickness definition by getting mean value computation.
The CT data that are used for model training are by the live body collection, position in world coordinate system has accurate correspondence by skull of rebuilding in the CT data and faceform, but we and do not know the skull model and faceform's grid in corresponding relation between the summit, be difficult to obtain tissue thickness between the two by direct computing; The skull that needs in addition to carry out the reconstruction of cranium face uses 3-D scanning to obtain, owing to exist and to block that to be difficult to the skull model rebuild with CT the same complete, and we need calculate the mapping relations between the general skull model that obtains in skull model that scanning obtains and the CT data in the skull model; Facial reconstruction is only relevant with outer bone, and the skull model of rebuilding in the CT image is the geometry with thickness, need divest its bone internal layer.The problems referred to above can solve by the sampling of equal resolution.
The model two-dimensional parameterization is based on conic projection, at first circular cylindrical coordinate (the r of each point on the computation model i, θ i, z i), this is put and specifies radius r sPeriphery on coordinate just can be expressed as (r s* θ i, z i).In order to solve the topological inconsistent problem of skull and people's face geometric model, we carry out resampling to two models with identical resolution, and its corresponding radial and axial sampling rate is respectively N rAnd N z, thereby the space is divided for N r* N zIndividual little lattice calculate from the light at cylinder main shaft process sampling grid center and the intersection point of model, obtain intersection point set P Inter={ p i| p i∈ P M, P wherein MIt is the point of model surface.According to main shaft apart from antinode set P InterOrdering, the point of getting depth capacity is as current grid
Figure BSA00000137296400041
Sampled point, the value of the degree of depth of this point as this grid respective pixel.The model of sampling back depth map correspondence has the protruding characteristic of guarantor on the conic projection meaning.Sampling is exactly to calculate skull and the faceform degree of depth with respect to the sampling main shaft simultaneously with the resolution of artificial regulation.
Owing to around eye socket bone, nasal bone and cheekbone, there is the cavity on the skull, facial three-dimensional model in the nostril, the upperlip bonding station may lack data, just there is the cavity on the grid to skull and the sampling of people's face like this, can not directly carry out deep operations and obtain tissue thickness's data, thereby need carry out the filling-up hole operation depth-sampling figure.Native system adopts simple filling-up hole strategy, for the cavity in the depth map with the degree of depth average of non-empty pixel in its eight neighborhood the degree of depth as this point.
Figure BSA00000137296400042
D wherein cBe the depth value of current point, d CiIt is the depth value of non-empty pixel in pixel eight neighborhoods.The filling-up hole process prescription is as follows:
(a) the boundary pixel B{p in search cavity i;
(b) for B{p iIn pixel, use the corresponding degree of depth average of its non-NULL eight neighborhood territory pixels as its depth value;
(c) the boundary pixel B{p among the renewal figure i, if B{p iBe not empty, continue step (b).
By above-mentioned algorithm, use the depth value on empty border to obtain the degree of depth of all pixels in the cavity by interpolation.After sampling finishes, all formed cylindrical radial for skull and people's face and protected protruding sampling grid model, skull by equal resolution and the faceform according to pixels degree of depth subtract each other, and just can obtain tissue thickness and scheme.Then tissue thickness's average obtains in the sampling colony by calculating in general tissue thickness's definition.Because the unique point in the soft tissue chart of different people distributes and is incomplete same, can not directly carry out mean value computation, must carry out registration according to face and skeleton character to soft tissue chart earlier, we use radial primary function network that the figure of tissue thickness that obtains from the different numbers of people is carried out registration.
We use radial primary function network that the figure of facial tissue that is obtained by different number of people data is carried out registration, and then the definition of the tissue thickness of computer general.Every organization chart comprises a group of feature point P Lmi, wherein unique point is used
Figure BSA00000137296400051
Mark.Specify an organization chart as benchmark, its characteristic of correspondence point set is P Lm0, other the figure of tissue thickness feature point set is P Lmi, use the mapping between two non-homogeneous figure of tissue thickness of radial primary function network training, ignore the affined transformation part here,
Figure BSA00000137296400052
After the use radial primary function network was out of shape the figure of tissue thickness, parameter sampling grid originally was distorted, and in order to operate with the reference map respective pixel, need carry out resampling to the figure of tissue thickness after the distortion, the thickness d of new sampling point position I, jThickness by the grid vertex after the distortion around it
Figure BSA00000137296400053
On average obtain.
The figure of all tissue thicknesses is carried out just can being defined by following formula computer general tissue thickness after the corresponding registration,
d r , z g = 1 N TN Σ k = 1 N TM d r , z k ,
Wherein
Figure BSA00000137296400055
Be general organization chart respective pixel (r s* θ g, z g) the degree of depth, N TMBe the number of sample in the current class,
Figure BSA00000137296400056
Be that k opens organization chart at location of pixels (r s* θ k, z k) the degree of depth.Use identical calculating can from the CT data, obtain bone stretch-out view with general organization chart correspondence.
By above-mentioned steps, we can obtain the definition of people's softhearted tissue morphology of one group of parameter correspondence.
2, the automatic extraction of skull feature point set
For the facial soft tissue chart of registration preferably, need denser facial characteristics point set usually, if this feature point set is not only very dull fully by manual mark, also have in the process of a large amount of unique points of mark various mistakes to occur.We have proposed a feature point set definition mode based on the grid segmentation.The at first facial characteristics set of definition correspondence on the source and target depth map, this needs to determine on two width of cloth figure that by hand about 20 points are right, uses the Delaunay triangulation then, according to the unique point generation characteristic of correspondence grid of these definition.By on the source and target feature grid, carrying out identical triangle segmentation, just can on two depth maps, obtain increasing unique point.In each segmentation, be m if comprise leg-of-mutton number in the current feature grid, so once segmentation just can obtain m unique point.For the triangle v in the current grid 1v 2v 3, the vertex v that corresponding segmentation is added is defined as,
v=b 1v 1+b 2v 2+b 3v 3,(0<b i<1)
B wherein 1, b 2And b 3Be the BaryCentric coordinate of triangle inside, they all are positioned in [0,1] interval.All can add training data for the feature point set that obtains of segmentation each time and remove to generate new radial primary function network, calculate the deformation pattern of new radial primary function network generation and the error between the target image, if error is less than a threshold value, just stop segmentation, use the training data of current feature set, otherwise continue the segmentation feature grid as network.Error is defined as follows,
e = Σ i N r Σ j N z | d ij s ′ - d ij i | ,
Wherein
Figure BSA00000137296400062
Be the depth value of pixel correspondence in the source images of distortion back,
Figure BSA00000137296400063
It is the depth value of respective pixel among the target depth figure.
We rebuild the overall situation covering that is defined as facial soft tissue layer and unknown bone.Wherein need the accurate registration organized, this registration is by the control of the feature point set in the corresponding bone depth map.Characteristic distribution in the soft tissue chart after the distortion is consistent with feature on the unknown skull.The distortion and the registration that use unknown skull and train soft tissue chart with reference to the feature point set on the skull.Stack by tissue thickness just can obtain the face points cloud with unknown head correspondence.
3,, estimate people's organ morphology on the face from the bone local measurement of bottom based on the classification of the cranium face structure shape of ethnic group
Human anatomy and science of heredity rule are the scientific theory bases of cranium face recovery technique: the recovery of cranium face is to infer according to the morphosis information of people's skull self to recover the dead's looks before death, so people's physique structure rule and people's law of development is one of foundation of this technology.Embody by general character and two aspects of individual character: the one, people's form has the general character rule, as the structure of people's head skeleton and soft tissue and to grow be clocklike, everybody unanimity.Skull is the framework core that people's facial feature constitutes, and plays the major decision effect, and on face and the corresponding site of other soft tissue attached to skull, the observations of morphosis aspect and measured data values are all within the specific limits.Adult's skull form is constant, head part's soft tissue thickness, and except that the fat or thin variation of cheek soft tissue individual was big, all the other positions were more constant, and difference is also less between the individual.The 2nd, people's form has the individual character rule, the people is because of race, age, or sex, morphosis difference, though there is the morphological differences on the individuality, but be familiar with in the middle of can getting back to the general character rule, as people's ethnic group, age, gender differences, the characteristics that reflect on skull are identical.People's skull structure and soft tissue growth change a guy's characteristics, constitute independent individual character, but this morphological differences is subjected to the influence and the restriction of the physique structure of skull because of it, formation corresponding structure relation, this structural relation is clocklike, that is to say that what kind of person shape of face and face which type of bone framework just produce, skeleton characters such as corresponding nasal bone, apertura piriformis are arranged as the people of aquiline nose feature.Therefore the process of cognition of general character from people's morphosis and the dialectical rule of individual character, can infer looks shape with its corresponding position soft tissue according to the physique structure characteristics at any position on the skull, thereby reach the purpose of the soft tissue looks form characteristics of inferring whole skull.
In order to make people's face geometric properties personalization of reconstruction, in the high layer model of reconfiguration system, use radial primary function network (RBF) to estimate people's organ morphology on the face from the bone local measurement of bottom.Local people's face shape feature is used the depth map definition, and sets up the shape subspace by the PCA technology.
In the people's face and skull depth map that the CT data are obtained, local feature can use a changeable shape region representation.The sampled point degree of depth in the polytrope zone constitutes vector can represent local feature:
Figure BSA00000137296400071
Wherein
Figure BSA00000137296400072
It is sampled point
Figure BSA00000137296400073
The corresponding degree of depth, n RegIt is the number of sampled point.reg=eye,nose,mouth。We need determine the feature correspondence in the depth map, and an improved ICP algorithm is used for determining the correspondence of local feature figure.
Figure BSA00000137296400074
Wherein
Figure BSA00000137296400075
Point after the conversion The degree of depth in new depth map.d 0It is the degree of depth defined function among the template characteristic figure.
Figure BSA00000137296400077
It is one 2 * 3 transformation matrix.We use the L-BFGS-B algorithm to find the solution this large-scale nonlinear optimal problem, obtain the transformation matrix on each sampled point
Figure BSA00000137296400078
The dimension of regional area is 3 * n RegHigher relatively.We adopt pca method to carry out dimensionality reduction.
D { f , s } reg = D ‾ { f , s } reg + w { f , s } reg · U { f , s } reg .
It is the degree of depth average of local feature.
Figure BSA000001372964000711
It is the characteristic of correspondence vector.Behind the dimensionality reduction, the depth map of local feature
Figure BSA000001372964000712
The low dimensional vector that a correspondence is arranged
Figure BSA000001372964000713
The depth map of new local organs can use pivot
Figure BSA000001372964000714
Linear combination obtain.
In the depth map that we obtain from one group of CT data, study local organs form is at the mapping function on skull and people's face top layer.Here use the RBF regression function
y i = f ( x i ) = Σ j = 1 q α j ker ( x i , μ j ) ,
Existing skull and the definition of people's face local feature are used to learn regression coefficient and kernel function center, and we use thin plate spline function as kernel function.
ker(x i,μ j)=(x ij) 2log|x ij|.
By the minimization of energy function
Figure BSA000001372964000716
Just can find the solution the parameter in the RBF regression function.Obtain skull and the people Feature Mapping of local organs on the face by study, the form of local organs definition on the given skull just can obtain correspondence at people's characterizing definition of form on the face by mapping function.
4, three-dimensional cranium face is rebuild
Skull for parked cranium face, with its parametrization to two dimensional surface, analyze its thickness distribution characteristic, seek approximate tissue thickness's distributed model, and at the facial local feature shape training radial primary function network realization input skull form of particular anatomy and the conversion between the general soft tissue thickness distribution plan, the figure of tissue thickness after the conversion is added on the input skull, and the three-dimensional cranium face that uses two-dimentional depth image image addition just to finish the input bone is rebuild.
System uses spatial digitizer to obtain its surface geometry data to the skull of new input, and the employing resolution identical with general organization chart resamples to it, use on the input skull and general skull characteristic of correspondence point set is controlled the distortion of general organization chart as training data study radial primary function network, to obtain the figure of tissue thickness with target skull correspondence.In order to guarantee to have the sampling definition of homology, general bone stretch-out view is also used identical network be out of shape with organization chart.The degree of depth that tissue thickness figure after the distortion and skull stretch-out view superposition has just been obtained the faceform corresponding with the target skull defines.
Use feature point set in general skull stretch-out view and the new input skull stretch-out view as the training data of radial primary function network, added the third dimension in the skull training data, promptly depth dimension is controlled the distortion of general skull, and its feature point set is defined as,
P lm = { p lm k ( r s × θ lm k , z lm k , d lm k ) } ,
At first calculate the weight matrix w of radial primary function network iIf the quantity of the feature point set that is used to train is N, we can obtain a W so 3 * NWeight matrix, obtain the submatrix W that corresponding cylinder launches the radial and axial bidimensional in the coordinate simultaneously [2 * N], because general tissue thickness remains unchanged, only need the registration facial characteristics in deformation process, thereby right to use matrix W [2 * N]Calculate the destination organization thickness chart; Then use W for skull 3 * N, promptly comprise depth dimension information, so just can be so that general skull model and the target skull model geometric similarity after the distortion.
P skull t = W 3 × N H ( P skull g ) ,
Wherein Be the stretch-out view pixel collection of distortion back and target skull shape approximation,
Figure BSA00000137296400084
It is the collection of pixels of general skull stretch-out view.
P tissue t = W 2 × N H ( P tissue g ) ,
Wherein Be destination organization thickness stretch-out view pixel set (r k* θ k, z k), It is the general plain set of unfolded image of organizing.
After obtaining the deformation result of general tissue and bone, just can obtain the degree of depth stretch-out view of corresponding target skull, obtain its three-dimensional model simultaneously by the stack of simple image pixel.
P face=P skull+P tissue
Wherein
Figure BSA00000137296400091
Definition is people's face, the collection of pixels of the degree of depth stretch-out view of skull or tissue, and its pixel coordinate comprises cylindrical radial, axial coordinate and the corresponding degree of depth.
In order to make reconstructed results personalized more, system allows reconstructed results is carried out mutual local deformation, by user's designated local region and define the control point set of this region shape, realize three-dimensional face model local deformation by this point set training radial primary function network.
Zone definitions on the three-dimensional model is the vertex set that communicates with each other
Figure BSA00000137296400092
In order to realize Deformation control, need the control point set of this region shape of specified control
Figure BSA00000137296400093
Specifically describe as follows:
(a)
Figure BSA00000137296400094
: be defined as the point set in i zone, this point set is the possible affected surface mesh summit of institute in the region deformation;
(b) : be defined as the border vertices in zone, it comprises a summit at least not at current region in eight neighborhoods of sampling stretch-out view
Figure BSA00000137296400096
In;
(c)
Figure BSA00000137296400097
: the shape control point set of regional i, this point set can provide by manual interaction, for the obvious part of feature, can obtain unique point by inspiring principle to search for automatically, and for example for nose, prenasale is corresponding to the point of attitude calibration occiput model forefront.
Use the training of Region control point set and frontier point to practice radial primary function network, in order to the distortion of control regional area.The input of training data is the preceding control point set of distortion and the union P of border point set In=C i r∪ B i r, training data is output as the control point set after the distortion and the union P of border point set Out=C i r' ∪ B i r, keep the border constant before and after distortion, force limit deformation at a regional area like this.Here consider the affined transformation of control point set, add by polynomial expression p at the distortion front and back position iLinear combination represent conversion such as rotation, translation, convergent-divergent,
Figure BSA00000137296400098
Original state D for local deformation 0With final state D 1Between shape can obtain by interpolation, adopt linear interpolation to obtain state in the middle of it in the native system.Interpolation parameter c (c ∈ [0,1]) is set, for original state c=0, for final state c=1, state D in the middle of it 1Calculate by following formula.
D t=D 0+c(D 1-D 0)。
5, face texture mapping
In order to have the stronger sense of reality behind the modeling rendering that makes reconstruction, added necessary face texture in the system, we use the synthetic facial complete texture maps of orthogonal pictures.Wherein two photos are taken from the dead ahead and the side of people's face respectively, and wherein camera does not need to demarcate, by the texture coordinate of choosing computational geometry model correspondence of unique point.Use mirror image to obtain an other side of its symmetry for the side photo, abutment joint obtains a complete face texture figure after doing smoothing processing.The texture coordinate that needs the correspondence of computational geometry model vertices in texture maps, the two-dimensional development figure with geometric model is mapped on the corresponding texture maps by the feature correspondence.The unique point M on the texture maps is chosen in the use manual interaction t, for unique point M in the geometric model stretch-out view gCan be undertaken by the image outline extractive technique is auxiliary.
Set up the model depth map P under the unique point control dTexture maps P with correspondence tBetween mapping just be converted into the process of seeking suitable interpolation operator F, wherein P t=F (P d).This interpolation operator need satisfy the constraint condition M of unique point t=F (M d).We use radial basis function as interpolating function, and wherein basis function is chosen the many quadratic functions of Hardy.
We have proposed the cranium face reconstruction technique based on the facial soft tissue thickness figure of two dimension, by conic projection the three-dimensional model parametrization is arrived two dimensional surface, extract complete facial soft tissue layer, use the overlap-add operation according to pixels in the image to realize rebuilding based on the skull faceform.Utilize radial primary function network to control not the bone of homology and the tissue thickness corresponding with importing skull obtained in the distortion of tissue, obtain and import the faceform of skull correspondence then by simple stack.System allows the faceform who generates is carried out the local deformation operation, and has constructed the mapping function of local organs shape between skull and people's face, is used to generate personalized faceform.
Utilize method of the present invention, can fast and effeciently carry out three-dimensional cranium face and rebuild unknown skull; Adopted the facial soft tissue thickness definition of from the CT training data, obtaining; Input skull form is analyzed, introduced the automatic acquisition algorithm of feature point set, and based on radial primary function network the soft tissue morphological transformation of training in the storehouse become and the consistent form of input skull, obtain the three-dimensional face form of reconstruction by the stack of the thickness on two parameter planes; And the mapping function that has provided local form editor and local feature between skull and the people's face can calculate personalized facial shape.Compare with traditional algorithm, can effectively overcome the missing reconstruction details in the classic method, the problem of lacking individuality feature.
Above embodiment only is used to illustrate the present invention; and be not limitation of the present invention; the those of ordinary skill in relevant technologies field; under the situation that does not break away from the spirit and scope of the present invention; can also make various variations and modification; therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (2)

1. the three-dimensional cranium face restored method based on Chinese's full-face structure graphic data is characterized in that, may further comprise the steps:
(1) extracts the middle soft tissue layer of holocephatic, launch it is projected two dimensional surface, use the depth map of two dimension to represent this organized layer by cylinder;
(2) utilize the automatic extractive technique of feature templates to obtain feature mark in the skull depth map;
(3) use radial primary function network to estimate local feature from the bone local measurement of bottom;
(4) for the skull of parked cranium face, with its parametrization to two dimensional surface, analyze its morphological feature, seek approximate tissue thickness's distributed model, at the radial primary function network realization input skull form of particular anatomy character shape training and the conversion between the general soft tissue thickness distribution plan, tissue thickness's distribution plan after the conversion is added on the input skull, and the three-dimensional cranium face that uses two-dimentional depth map addition to finish the input bone is rebuild;
(5), use the synthetic facial complete texture maps of orthogonal pictures, and the colour of skin, the hair style of restoring portrait are played up by face texture mapping.
2. according to the described three-dimensional cranium face restored method of claim 1, it is characterized in that described step (1) further may further comprise the steps based on Chinese's full-face structure graphic data:
(a) from the CT view data, generate skull and faceform;
(b) with the skull of correspondence and faceform's parametrization to two dimensional surface, use specified resolution to carry out double sampling, people's face and the skull depth map with respect to the sampling central shaft is obtained in sampling;
(c) skull by correspondence and the figure of facial tissue subtract each other and obtain soft tissue chart;
(d) use radial basis function that all organization charts in the same class are carried out registration according to its feature, obtain general soft tissue thickness definition by getting mean value computation.
CN2010101806286A 2010-05-18 2010-05-18 Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people Pending CN101882326A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010101806286A CN101882326A (en) 2010-05-18 2010-05-18 Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010101806286A CN101882326A (en) 2010-05-18 2010-05-18 Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people

Publications (1)

Publication Number Publication Date
CN101882326A true CN101882326A (en) 2010-11-10

Family

ID=43054336

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010101806286A Pending CN101882326A (en) 2010-05-18 2010-05-18 Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people

Country Status (1)

Country Link
CN (1) CN101882326A (en)

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107678153A (en) * 2017-10-16 2018-02-09 南京亘瑞医疗科技有限公司 Fibre bundle image processing method and device
CN107993277A (en) * 2017-11-28 2018-05-04 河海大学常州校区 Damage location artificial skelecton patch formation model method for reconstructing based on priori
CN108205806A (en) * 2016-12-20 2018-06-26 北京大学 A kind of automatic analytic method of pyramidal CT image three-dimensional cranio-orbital tumor
CN108257136A (en) * 2018-02-09 2018-07-06 天津海达奥普光电技术股份有限公司 A kind of image partition method of corn seed Shape Feature Extraction
CN108346175A (en) * 2018-02-06 2018-07-31 腾讯科技(深圳)有限公司 A kind of facial image restorative procedure, device and storage medium
CN109118455A (en) * 2018-09-14 2019-01-01 北京师范大学 A kind of ancient human's skull cranium face interactive restoration method based on the distribution of modern's soft tissue
CN109528323A (en) * 2018-12-12 2019-03-29 上海牙典软件科技有限公司 A kind of orthodontic procedure and device based on artificial intelligence
CN109615702A (en) * 2018-10-23 2019-04-12 深圳市华讯方舟太赫兹科技有限公司 Grid filling-up hole method, image processing equipment and the device with store function
CN109636910A (en) * 2018-12-18 2019-04-16 四川大学 A kind of cranium face restored method generating confrontation network based on depth
CN110264562A (en) * 2019-05-23 2019-09-20 吴薇 Skull model characteristic point automatic calibration method
CN110335358A (en) * 2019-06-18 2019-10-15 大连理工大学 The personalized deformation method of deformable Digital Human anatomical model
CN110390148A (en) * 2019-07-08 2019-10-29 浙江工业大学 A kind of incidence soft tissue defects reconstruction operations template design method and surgical template based on curved surface features classification processing and expansion
CN111243413A (en) * 2020-03-06 2020-06-05 吉林大学 Modeling method and teaching system for facial anatomy teaching
CN111460899A (en) * 2020-03-04 2020-07-28 达理 Soft and hard tissue feature topology identification and facial deformation prediction method based on deep learning
WO2020206135A1 (en) * 2019-04-02 2020-10-08 The Methodist Hospital System Image-based methods for estimating a patient-specific reference bone model for a patient with a craniomaxillofacial defect and related systems
CN112070900A (en) * 2020-09-18 2020-12-11 北京交通大学 Modeling method of head skull morphological model of 3-10 year old child
US10937250B2 (en) 2016-12-06 2021-03-02 National University Of Singapore Methods of reconstructing skulls
WO2021210723A1 (en) * 2020-04-16 2021-10-21 이마고웍스 주식회사 Method and apparatus for automatically detecting feature points of three-dimensional medical image data by using deep learning
CN114445514A (en) * 2022-01-26 2022-05-06 四川大学 Template data generation and application method based on magnetic resonance scanning image

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070183653A1 (en) * 2006-01-31 2007-08-09 Gerard Medioni 3D Face Reconstruction from 2D Images
JP2007265396A (en) * 2006-03-29 2007-10-11 Mitsubishi Electric Research Laboratories Inc Method and system for generating face model
CN101339670A (en) * 2008-08-07 2009-01-07 浙江工业大学 Computer auxiliary three-dimensional craniofacial rejuvenation method
CN101383046A (en) * 2008-10-17 2009-03-11 北京大学 Three-dimensional reconstruction method on basis of image

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070183653A1 (en) * 2006-01-31 2007-08-09 Gerard Medioni 3D Face Reconstruction from 2D Images
JP2007265396A (en) * 2006-03-29 2007-10-11 Mitsubishi Electric Research Laboratories Inc Method and system for generating face model
CN101339670A (en) * 2008-08-07 2009-01-07 浙江工业大学 Computer auxiliary three-dimensional craniofacial rejuvenation method
CN101383046A (en) * 2008-10-17 2009-03-11 北京大学 Three-dimensional reconstruction method on basis of image

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
《15th IEEE International Conference on Image Processing》 20081015 Yuru Pei 等 Creating a Face Model from an Unknown Skull Based on the Tissue Map 第2088-2091页 1-2 , 2 *
《19th International Conference on Pattern Recognition》 20081211 Yuru Pei 等 Facial Feature Estimation from the Local Structural Diversity of Skulls , 2 *
《Proceedings of the Third International Conference on Image and Graphics》 20041231 Yuru Pei 等 Tissue Map based Craniofacial Reconstruction and Facial Deformation Using RBF Network 1-2 , 2 *
《中国国家图书馆特色资源(博士)》 20071231 裴玉茹 人脸形状分析和视频驱动三维语音动画研究 , 2 *

Cited By (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10937250B2 (en) 2016-12-06 2021-03-02 National University Of Singapore Methods of reconstructing skulls
CN108205806B (en) * 2016-12-20 2020-10-09 北京大学 Automatic analysis method for three-dimensional craniofacial structure of cone beam CT image
CN108205806A (en) * 2016-12-20 2018-06-26 北京大学 A kind of automatic analytic method of pyramidal CT image three-dimensional cranio-orbital tumor
US11525996B2 (en) 2017-10-16 2022-12-13 Suzhou Microview Medical Technologies Co., Ltd. Fiber bundle image processing method and apparatus
CN107678153A (en) * 2017-10-16 2018-02-09 南京亘瑞医疗科技有限公司 Fibre bundle image processing method and device
CN107993277A (en) * 2017-11-28 2018-05-04 河海大学常州校区 Damage location artificial skelecton patch formation model method for reconstructing based on priori
CN107993277B (en) * 2017-11-28 2019-12-17 河海大学常州校区 priori knowledge-based reconstruction method of artificial bone repair model of damaged part
CN108346175A (en) * 2018-02-06 2018-07-31 腾讯科技(深圳)有限公司 A kind of facial image restorative procedure, device and storage medium
CN108346175B (en) * 2018-02-06 2023-10-20 腾讯科技(深圳)有限公司 Face image restoration method, device and storage medium
CN108257136A (en) * 2018-02-09 2018-07-06 天津海达奥普光电技术股份有限公司 A kind of image partition method of corn seed Shape Feature Extraction
CN109118455B (en) * 2018-09-14 2021-12-10 北京师范大学 Ancient human skull craniofacial interactive restoration method based on modern soft tissue distribution
CN109118455A (en) * 2018-09-14 2019-01-01 北京师范大学 A kind of ancient human's skull cranium face interactive restoration method based on the distribution of modern's soft tissue
CN109615702B (en) * 2018-10-23 2023-09-22 深圳市重投华讯太赫兹科技有限公司 Grid hole filling method, image processing equipment and device with storage function
CN109615702A (en) * 2018-10-23 2019-04-12 深圳市华讯方舟太赫兹科技有限公司 Grid filling-up hole method, image processing equipment and the device with store function
CN109528323B (en) * 2018-12-12 2021-04-13 上海牙典软件科技有限公司 Orthodontic method and device based on artificial intelligence
CN109528323A (en) * 2018-12-12 2019-03-29 上海牙典软件科技有限公司 A kind of orthodontic procedure and device based on artificial intelligence
CN109636910A (en) * 2018-12-18 2019-04-16 四川大学 A kind of cranium face restored method generating confrontation network based on depth
WO2020206135A1 (en) * 2019-04-02 2020-10-08 The Methodist Hospital System Image-based methods for estimating a patient-specific reference bone model for a patient with a craniomaxillofacial defect and related systems
CN110264562A (en) * 2019-05-23 2019-09-20 吴薇 Skull model characteristic point automatic calibration method
CN110264562B (en) * 2019-05-23 2023-06-27 沈阳医学院 Automatic calibration method for feature points of skull model
CN110335358A (en) * 2019-06-18 2019-10-15 大连理工大学 The personalized deformation method of deformable Digital Human anatomical model
CN110390148A (en) * 2019-07-08 2019-10-29 浙江工业大学 A kind of incidence soft tissue defects reconstruction operations template design method and surgical template based on curved surface features classification processing and expansion
CN110390148B (en) * 2019-07-08 2023-09-26 浙江工业大学 Design method of head and neck soft tissue defect reconstruction operation template and operation template
CN111460899B (en) * 2020-03-04 2023-06-09 达理 Soft and hard tissue characteristic topology identification and facial deformation prediction method based on deep learning
CN111460899A (en) * 2020-03-04 2020-07-28 达理 Soft and hard tissue feature topology identification and facial deformation prediction method based on deep learning
CN111243413B (en) * 2020-03-06 2021-07-02 吉林大学 Modeling method and teaching system for facial anatomy teaching
CN111243413A (en) * 2020-03-06 2020-06-05 吉林大学 Modeling method and teaching system for facial anatomy teaching
WO2021210723A1 (en) * 2020-04-16 2021-10-21 이마고웍스 주식회사 Method and apparatus for automatically detecting feature points of three-dimensional medical image data by using deep learning
CN112070900A (en) * 2020-09-18 2020-12-11 北京交通大学 Modeling method of head skull morphological model of 3-10 year old child
CN112070900B (en) * 2020-09-18 2023-11-24 北京交通大学 Modeling method of 3-10 years old child head skull morphological model
CN114445514A (en) * 2022-01-26 2022-05-06 四川大学 Template data generation and application method based on magnetic resonance scanning image
CN114445514B (en) * 2022-01-26 2023-04-07 四川大学 Template data generation and application method based on magnetic resonance scanning image

Similar Documents

Publication Publication Date Title
CN101882326A (en) Three-dimensional craniofacial reconstruction method based on overall facial structure shape data of Chinese people
CN106023288B (en) A kind of dynamic scapegoat's building method based on image
Khamis et al. Learning an efficient model of hand shape variation from depth images
CN101755288B (en) Method and apparatus for realistic simulation of wrinkle aging and de-aging
De Aguiar et al. Automatic conversion of mesh animations into skeleton‐based animations
CN101339670B (en) Computer auxiliary three-dimensional craniofacial rejuvenation method
US9898848B2 (en) Co-registration—simultaneous alignment and modeling of articulated 3D shapes
Cheng et al. Parametric modeling of 3D human body shape—A survey
CN104794722A (en) Dressed human body three-dimensional bare body model calculation method through single Kinect
CN109377557A (en) Real-time three-dimensional facial reconstruction method based on single frames facial image
CN104346824A (en) Method and device for automatically synthesizing three-dimensional expression based on single facial image
CN109118455B (en) Ancient human skull craniofacial interactive restoration method based on modern soft tissue distribution
Deng et al. A regional method for craniofacial reconstruction based on coordinate adjustments and a new fusion strategy
Shui et al. A computerized craniofacial reconstruction method for an unidentified skull based on statistical shape models
Oliveira-Santos et al. 3D face reconstruction from 2D pictures: first results of a web-based computer aided system for aesthetic procedures
CN102034272A (en) Generating method of individualized maxillofacial soft tissue hexahedral mesh
Abate et al. FACES: 3D FAcial reConstruction from anciEnt Skulls using content based image retrieval
Zhang et al. Anatomy-based face reconstruction for animation using multi-layer deformation
Liu et al. Three-dimensional cartoon facial animation based on art rules
Liu et al. RAS: A Data‐Driven Rigidity‐Aware Skinning Model For 3D Facial Animation
CN109859306A (en) A method of extracting manikin in the slave photo based on machine learning
Subsol et al. Automatic 3D facial reconstruction by feature-based registration of a reference head
Patterson et al. Landmark-based re-topology of stereo-pair acquired face meshes
Fang et al. A modeling method for the human body model with facial morphology
Kermi et al. 3D-Computerized facial reconstructions from 3D-MRI of human heads using deformable model approach

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C12 Rejection of a patent application after its publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20101110