CN106570450B - The detection method and its system of prenasale on three-dimensional face based on curvature distribution - Google Patents
The detection method and its system of prenasale on three-dimensional face based on curvature distribution Download PDFInfo
- Publication number
- CN106570450B CN106570450B CN201610809148.9A CN201610809148A CN106570450B CN 106570450 B CN106570450 B CN 106570450B CN 201610809148 A CN201610809148 A CN 201610809148A CN 106570450 B CN106570450 B CN 106570450B
- Authority
- CN
- China
- Prior art keywords
- curvature
- vertex
- vertex set
- face
- histogram
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention is suitable for three-dimensional face identification technical field, provide the detection method and its system of the prenasale based on curvature distribution on a kind of three-dimensional face, the grid surface that the detection method includes the following steps: A, constructs three-dimensional face, the curvature and curvature distribution on the vertex on the network curved surface are calculated, and draws the corresponding curvature histogram of the curvature;B, the vertex according to the connectivity pair of the curvature distribution and the curvature histogram carries out triple filtering processings, obtains filtered vertex set;C, it is fitted face principal plane on the filtered vertex set, and the vertex in the normal direction obtained in plane domain on the face principal plane is positioned as the prenasale.Whereby, the present invention, which realizes, keeps the detection of the prenasale of three-dimensional face more easy.
Description
Technical field
The present invention relates to the noses based on curvature distribution in three-dimensional face identification technical field more particularly to a kind of three-dimensional face
The detection method and its system of cusp.
Background technique
With the development of three-dimensional automatically scanning and modeling technique, the acquisition of three-dimensional face is also increasingly mature.Three-dimensional face is not
It is influenced by factors such as posture, illumination, is widely used in fields such as social safety check, court's identifications.It is being that three-dimensional face is swept
During retouching, the three-dimensional point cloud of acquisition is not simple face face area, and also comprising parts such as neck, hair and clothes, this is mentioned
Three-dimensional face features are taken to bring very big adverse effect.In all organs of face, nasal region is the most prominent, and geometrical characteristic is most
To be obvious, very big function served as bridge whether is played in two-dimension human face and three-dimensional face detection.
Three-dimensional face detection research has had the research history of more than ten years, Colombo etc. [Colombo2006]
(Alessandro Colombo,Claudio Cusano,Raimondo Schettini.3D face detection using
Curvature analysis.Pattern Recognition 39 (2006) 444-455.) height is calculated on rangeimage
This curvature and average curvature are obtained the candidate item of Vitrea eye and nasal region by filtering, are learnt using HK classifier, by all times
Favored area is divided into four classes, concave region, convex domain, two breeds of horses saddle region, then detects human face region by priori knowledge and concavity and convexity.
Bevilacqua etc. [Bevilacqua2008] (Vitoantonio Bevilacqua, Pasquale Casorio, Giuseppe
Mastronardi.Extending Hough Transform to a Points’Cloud for 3D-Face Nose-Tip
Detection.Lecture Notes in Computer Science,2008,vol 5227,pp 1200-1209.)
It is detected on three-dimensional point cloud using hough transform, detects prenasale position with ballot method.Yang etc. [Yang2009]
(Yang Jun,Liao Zhi-Wu,etal.A Method for Robust Nose Tip Location across Pose
Variety in 3D Face Data.Proceeding on International Asia Conference on
Informatics in Control, Automation and Robotics, 2009, pp114-117.) first on three-dimensional face
The candidate point for choosing several maximum normal direction, three-dimensional face is rotated, so that nose is aligned in forward direction with pca method, building is horizontal
The curve in direction, the curve with maximum are confirmed as being decided to be prenasale by the curve of prenasale, maximum point.
The grid median filter method of Bagchi etc. [Bagchi2012] proposition Weight on rangeimage to three-dimensional face into
Row filtering processing, this processing can enhance the brightness of nasal region, and most bright point is defined as prenasale.Cai etc. [Cai2012]
(Parama Bagchi,Debotosh Bhattacharjee,etal.A novel approach for nose tip
detection using smoothing by weighted median filtering applied to 3D face
images in variant poses.Proceedings of the International Conference on
Pattern Recognition, Informatics and Medical Engineering, 2012, pp272-277.) it utilizes
Average curvature detects prenasale position, they have found that the maximum region of face average curvature is located near nose, and using in this
Information detection three-dimensional prenasale.But this discovery is in non-simple faceform always not total establishment, and maximum point can also
Can there are eye and lip.Priya etc. [Priya2012] (M.V.Priya, R.Narayanan, etal.A Combined
Approach for Face Detection with 3D Mesh and Eigen Face.Procedia Engineering
38 (2012) 298-305.) method of eigenface is proposed for detection of three dimensional face, they are corresponding by calculating three-dimensional face
Two dimensional image feature vector, the covariance matrix for being averaging vector, and establishing vertex decomposes before covariance matrix obtains n most
Big characteristic value and feature vector, decomposes three-dimensional face using these feature vectors, obtains the spectrum signature of three-dimensional face,
It is also required to be decomposed in detection, according to threshold comparison spectrum signature.Mukherjee etc. [Mukherjee2014] is utilized
Curvature and HK classifier carry out feature extraction prenasale.This method is existing to project to curved surface for curvature, obtains curvature image, and with bent
Curvature image block is defined three kinds of templates by rate, then is handled with morphological method, and prenasale is obtained.KHADHRAOUI etc.
[KHADHRAOUI2014](Taher KHADHRAOUI,Faouzi BENZARTI,etal.A Novel Approach to
Nose-tip and Eye-corners Detection using HK-classification in case of 3D Face
Analysis.Proceedings on World Congress on Computer Applications and
Information Systems (WCCAIS), 2014, pp1-4.) it is also the method detection people that curvature is combined with HK classifier
Face, with [Mukherjee2014] (Debashis Mukherjee, Debotosh Bhattacharjee, etal.Curvature
Based Localization of Nose Tip Point for Processing 3D-Face from Range
Images.Proceedings on International Conference on Communication and Signal
Processing, 2014, pp046-050.) the difference is that, they directly classify on three-dimensional face.Segundo
Deng [Segundo2014] (M.Pamplona Segundo, L.Silva, etal.Orthogonal projection images
For 3D face detection.Pattern Recognition Letters50 (2014) 72-81.) divided with boost cascade
Class device detects the face in rangeimage, and this method needs to be standardized the size of face, it is also necessary to obtain
The rangeimage in the multiple rectangular projection directions of face extracts characteristic value to multiple rangeimage, is learnt, finally carried out
Three-dimensional face detection.
In summary, existing, in actual use, it is clear that there is inconvenient and defect, so it is necessary to be improved.
Summary of the invention
For above-mentioned defect, the purpose of the present invention is to provide the prenasales based on curvature distribution on a kind of three-dimensional face
Detection method and its system so that the detection of the prenasale of three-dimensional face is more easy.
To achieve the goals above, the present invention provides a kind of detection side of the prenasale based on curvature distribution on three-dimensional face
Method, the detection method include the following steps:
A, the grid surface for constructing three-dimensional face, calculates the curvature and curvature distribution on the vertex on the network curved surface, and
Draw the corresponding curvature histogram of the curvature;
B, the vertex according to the connectivity pair of the curvature distribution and the curvature histogram carries out triple filtering processings,
Obtain filtered vertex set;
C, it is fitted face principal plane on the filtered vertex set, and plane area will be obtained on the face principal plane
The vertex in normal direction in domain is positioned as the prenasale.
According to the detection method, the step A includes:
A1, the image of the three-dimensional face of acquisition is initialized, constructs the grid surface of the three-dimensional face images;
A2, denoising is carried out to the grid surface, and calculate vertex on the network curved surface Gaussian curvature,
Average curvature and two principal curvatures
A3, corresponding curvature distribution is calculated using the Gaussian curvature or average curvature or two principal curvatures, and drawn
Make corresponding curvature histogram.
According to the detection method, the step B includes:
The upper threshold value th_t and lower threshold value th_b of B1, the setting curvature histogram, according to the curvature distribution to described
Curvature filtering;
B2, the vertex is divided by three classes according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram, wherein
Vertex set less than the lower threshold value th_b is vertex set VS0, is in vertex between the lower threshold value th_b and upper threshold value th_t
Integrating the vertex set as vertex set VS1, greater than the upper threshold value th_t is vertex set VS2;
B3, in the vertex set VS0 and vertex set VS2, according to the connectivity of the curvature histogram search connection area
Domain is filtered the connection component in the connection region, and only retains the maximum connection component in the connection component
Vertex set in the maximum connection component VS3 is labeled as VM by VS3;
B4, it is filtered again in the vertex set VM, from the top filtered out in the vertex set VM in the vertex set VS2
Point set VS3, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and calculate the vertex set
The bounding box of VS4 is labeled as bounding box Box0.
According to the detection method, the step C includes:
C1, it is fitted face principal plane p0 on the vertex set VS4, and takes the plane domain in the bounding box Box0
p1。
C2, on the plane domain in p1, most preceding vertex is set to the prenasale of face in face p1 normal direction of making even.
According to the detection method, the curvature histogram are as follows:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the grade of histogram
Number, len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v
(ci), v is the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, put down
Equal one of curvature and two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viIt is
I vertex, i are the serial number on i-th of vertex.
In order to realize another goal of the invention of the invention, the present invention also provides curvature distribution is based on a kind of three-dimensional face
Prenasale detection system, the detection system includes:
Computing module, for constructing the grid surface of three-dimensional face, calculate the vertex on the network curved surface curvature and
Curvature distribution, and draw the corresponding curvature histogram of the curvature;
Filtering module carries out three for the vertex according to the connectivity pair of the curvature distribution and the curvature histogram
It is filtered again, obtains filtered vertex set;
Module is obtained, for being fitted face principal plane on the filtered vertex set, and by the face principal plane
The vertex in normal direction in upper acquisition plane domain is positioned as the prenasale.
According to the detection system, the computing module includes:
Submodule is constructed, the image for the three-dimensional face to acquisition initializes, and constructs the three-dimensional face images
Grid surface;
First computational submodule for carrying out denoising to the grid surface, and calculates on the network curved surface
Vertex Gaussian curvature, average curvature and two principal curvatures
Second computational submodule, for calculating and corresponding to using the Gaussian curvature or average curvature or two principal curvatures
Curvature distribution, and draw corresponding curvature histogram.
According to the detection system, the filtering module includes:
First filtering submodule, for setting the upper threshold value th_t and lower threshold value th_b of the curvature histogram, according to institute
Curvature distribution is stated to filter the curvature;
Classification submodule, for being divided the vertex according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram
At three classes, wherein the vertex set less than the lower threshold value th_b is vertex set VS0, is in the lower threshold value th_b and upper threshold value
Vertex set is vertex set VS1 between th_t, and the vertex set greater than the upper threshold value th_t is vertex set VS2;
Second filtering submodule, in the vertex set VS0 and vertex set VS2, according to the connection of the curvature histogram
Property search connection region, the connection component in the connection region is filtered, and only retains in the connection component most
Vertex set in the maximum connection component VS3 is labeled as VM by big connection component VS3;
Third filters submodule, filters again in the vertex set VM, filters out the top from the vertex set VM
Vertex set VS3 in point set VS2, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and count
The bounding box of the vertex set VS4 is calculated, bounding box Box0 is labeled as.
According to the detection system, the acquisition module includes:
It is fitted submodule, for being fitted face principal plane p0 on the vertex set VS4, and is taken in the bounding box Box0
Plane domain p1.
Prenasale determines submodule, in p1, most preceding vertex to be fixed in face p1 normal direction of making even on the plane domain
For the prenasale of face.
According to the detection system, the curvature histogram are as follows:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the grade of histogram
Number, len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v
(ci), v is the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, put down
Equal one of curvature and two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viIt is
I vertex, i are the serial number on i-th of vertex.
Grid surface by constructing three-dimensional face of the invention, calculates the curvature and song on the vertex on the network curved surface
Rate distribution, and draw the corresponding curvature histogram of the curvature;Then according to the curvature distribution and the curvature histogram
Vertex described in connectivity pair carries out triple filtering processings, obtains filtered vertex set;Finally, in the filtered vertex set
Upper fitting face principal plane, and the vertex in the normal direction obtained in plane domain on the face principal plane is positioned as the nose
Cusp.Thus, it is possible to keep the detection of prenasale more easy, do not need specifically to project, does not need to be aligned, directly be counted
It calculates, the time complexity of algorithm is low, O (nlogn), for four kinds of curvature, Gaussian curvature, average curvature and maximum principal curvatures effect
Fruit is preferable.
Detailed description of the invention
Fig. 1 is the structure of the detection system of the prenasale on three-dimensional face provided in an embodiment of the present invention based on curvature distribution
Figure;
Fig. 2 is the structure of the detection system of the prenasale on three-dimensional face provided in an embodiment of the present invention based on curvature distribution
Figure;
Fig. 3 A is that Gaussian curvature provided in an embodiment of the present invention calculates schematic diagram;
Fig. 3 B is that average curvature provided in an embodiment of the present invention calculates schematic diagram;
Fig. 4 A is the Voronoi area schematic diagram at vertex P provided in an embodiment of the present invention;
Fig. 4 B is the Voronoi area schematic diagram at vertex P provided in an embodiment of the present invention;
Fig. 4 C is the Voronoi area schematic diagram at vertex P provided in an embodiment of the present invention;
Fig. 5 A is original face schematic diagram provided in an embodiment of the present invention;
Fig. 5 B is the average curvature distribution schematic diagram on three-dimensional face surface provided in an embodiment of the present invention;
Fig. 5 C is the curvature distribution schematic diagram after bandpass filtering provided in an embodiment of the present invention;
Fig. 5 D is the maximum connection component schematic diagram of curvature distribution provided in an embodiment of the present invention;
Fig. 5 E is face principal plane schematic diagram provided in an embodiment of the present invention;
Fig. 5 F is the principal plane of original face provided in an embodiment of the present invention
Fig. 5 G is prenasale detection result schematic diagram provided in an embodiment of the present invention;
Fig. 6 is the detection method process of the prenasale on three-dimensional face provided in an embodiment of the present invention based on curvature distribution
Figure;
Fig. 7 is the detection method process of the prenasale on three-dimensional face provided in an embodiment of the present invention based on curvature distribution
Figure.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Referring to Fig. 1, in the first embodiment of the present invention, the nose based on curvature distribution on a kind of three-dimensional face is provided
The detection system 100 of point, which is characterized in that detection system 100 includes:
Computing module 10 calculates the curvature on the vertex on the network curved surface for constructing the grid surface of three-dimensional face
And curvature distribution, and draw the corresponding curvature histogram of the curvature;
Filtering module 20 is carried out for the vertex according to the connectivity pair of the curvature distribution and the curvature histogram
Triple filtering processings obtain filtered vertex set;
Module 30 is obtained, for being fitted face principal plane on the filtered vertex set, and the face master is put down
The vertex obtained in the normal direction in plane domain on face is positioned as the prenasale.
In this embodiment, due to hardly resulting in simple three-dimensional face in three-dimensional human face scanning, and three-dimensional face
Posture is not known simultaneously, how in the face point cloud of different postures detection of three dimensional face, what feature can stably obtain face area with
It is critical issue, and the technical solution of the embodiment mainly solves the problems, such as to detect prenasale on three-dimensional face.Specifically, first
First computing module 10 constructs the grid surface of three-dimensional face, calculates the curvature and curvature distribution on the vertex on the network curved surface,
And draw the corresponding curvature histogram of the curvature;Filtering module 20 is then according to the curvature distribution and the curvature histogram
Vertex described in connectivity pair carries out triple filtering processings, obtains filtered vertex set;Module 30 is finally obtained in the filtering
It is fitted face principal plane on vertex set afterwards, and the vertex in the normal direction obtained in plane domain on the face principal plane is determined
Position is the prenasale.I.e. geometrically, prenasale is point most outstanding in positive face direction, can make full use of this priori features
Realize this detection to prenasale.Distribution of the curvature on face especially feature, in the gentle region of surface transfer, it changes very
It is small, and it is larger in the violent region transformation of surface transfer, such as nasal region, Vitrea eye and mouth area.Face area is detected using the distribution of curvature.
, not there is merely human face data in the three-dimensional face obtained with spatial digitizer, also have hair, neck, shoulder etc., Ke Nengye
There are the images of clothing, are found by verifying, and the distribution of the larger and smaller curvature of nasal region Vitrea eye and mouth area is with uniformity and connects
The general character, and the region occupied is maximum, and face area is different from other regions using this feature.
Referring to fig. 2, in the second embodiment of the present invention, the computing module 10 includes:
Submodule 11 is constructed, the image for the three-dimensional face to acquisition initializes, and constructs the three-dimensional face figure
The grid surface of picture;
First computational submodule 12 for carrying out denoising to the grid surface, and calculates the network curved surface
On vertex Gaussian curvature, average curvature and two principal curvatures
Second computational submodule 13, for using the Gaussian curvature or average curvature or two principal curvatures calculating pair
The curvature distribution answered, and draw corresponding curvature histogram.
The filtering module 20 includes:
First filtering submodule 21, for setting the upper threshold value th_t and lower threshold value th_b of the curvature histogram, according to
The curvature distribution filters the curvature;
Classify submodule 22, for according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram by the vertex
It is divided into three classes, wherein the vertex set less than the lower threshold value th_b is vertex set VS0, is in the lower threshold value th_b and upper-level threshold
Vertex set is vertex set VS1 between value th_t, and the vertex set greater than the upper threshold value th_t is vertex set VS2;
Second filtering submodule 23, in the vertex set VS0 and vertex set VS2, according to the company of the curvature histogram
The general character searches connection region, is filtered to the connection component in the connection region, and only retain in the connection component
Vertex set in the maximum connection component VS3 is labeled as VM by maximum connection component VS3;
Third filters submodule 24, filters again in the vertex set VM, filters out from the vertex set VM described
Vertex set VS3 in vertex set VS2, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and
The bounding box of the vertex set VS4 is calculated, bounding box Box0 is labeled as.
The acquisition module 30 includes:
It is fitted submodule 31, for being fitted face principal plane p0 on the vertex set VS4, and takes the bounding box Box0
In plane domain p1.
Prenasale determines submodule 32, is used on the plane domain in p1, most preceding vertex in face p1 normal direction of making even
It is set to the prenasale of face.
In this embodiment, the image initial of the three-dimensional face of 11 pairs of submodule acquisitions, including construction grid song are constructed
Face, the denoising of the first computational submodule 12 and calculating vertex curvature, including Gaussian curvature, average curvature and two principal curvatures.Second
Computational submodule 13 calculates curvature distribution.
Wherein, the calculation formula of curvature:
Gaussian curvature:
Average curvature: H (vi)=(∑j∈N(i)(cotαij+cotβij)(vi-vj))/Avoronoi (2)
Wherein v in formula (1)iFor i-th of vertex on curved surface, N is vertex PiAdjoining number of triangles, θjIt is j-th
Adjacent triangular apex viAngle, AvoronoiIt is all of its neighbor triangle in vertex viThe sum of the Voronoi diagram shape area at place,
See Figure of description 3A- Fig. 3 B.V in formula (2)i、vjI-th and j-th of vertex respectively on curved surface, N (i) are vertex vi
Adjacent vertex set, αijAnd βijTwo for side ij in two adjacent triangles are diagonal, AvoronoiFor all of its neighbor triangle
In vertex viThe sum of Voronoi diagram shape area at place, see Figure of description 4B, AvoronoiCalculating see Figure of description 4A
~Fig. 4 C and Figure of description 5A~Fig. 5 G.Specifically, being distributed and counting by average curvature since Fig. 5 A show original face
After calculation, the average curvature distribution on Fig. 5 B three-dimensional face surface is obtained;Curvature distribution after bandpass filtering later, such as Fig. 5 C institute
Show;Then the maximum connection component processing of curvature distribution is carried out again, as shown in Figure 5 D;Then face principal plane and original are successively extracted
The principal plane of beginning face finally detects prenasale as shown in Fig. 5 E and Fig. 5 F, is nose institute as shown in Fig. 5 G arrow
Show.
Principal curvatures:Its
Middle H (v) and G (v) are the average curvature and Gaussian curvature at vertex v respectively.
After completing curvature estimation, the first filtering submodule 21 sets the curvature histogram upper threshold value th_t and lower threshold value
Th_b filters curvature according to curvature distribution, and submodule 22 of classifying is divided into three classes according to th_t and th_b opposite vertexes, is less than th_b
Vertex set VS0, the vertex set VS1 and vertex set VS2 greater than th_t between th_b and th_t.Second filtering submodule 23 exists
Connection region is searched according to the connectivity of figure (what figure may I ask is) in vertex set VS0 and vertex set VS2, distich reduction of fractions to a common denominator amount carries out
Filtering, only retains maximum connection component, and vertex set is labeled as VM.Third filtering submodule 24 is filtered again in vertex set VM
Wave only retains the vertex set VS4 in VM in VS0 from the vertex set VS3 filtered out in VS2 in VM, and bounding box is labeled as
Box0。
Wherein, curvature histogram is defined as:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the grade of histogram
Number, len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v
(ci), v is the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, put down
Equal one of curvature and two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viIt is
I vertex, i are the serial number on i-th of vertex.
And vertex set is classified are as follows:
Vertex set VS0, VS0={ vi|Ci≤th_b},0≤i<N};
Vertex set VS1, VS1={ vi|th_b<Ci<th_t},0≤i<N};
Vertex set VS2, VS2={ vi|Ci≥th_t},0≤i<N};
Wherein CiFor vertex viThe curvature at place, N are vertex sum, and i is positive integer, and th_b, th_t are respectively curvature histogram
Lower threshold value and upper threshold value.
And it is connected to subset and largest connected subset are as follows:
VM=Max (len (VCi))
Wherein VCiFor i-th of connection subset, P (v on three-dimensional facek vh) it is vertex vk、vhBetween passage length, VM be most
Big connection subset, Max () are max function, and Len () calculates connected graph subset vertex quantity.
Maximum connection component filtering are as follows:
VS4=VM ∩ VS0;
Box0={ P0 P1},P0=(x0 y0 z0),P1=(x1 y1 z1);
x0=maxX (vi),x1=minX (vi),y0=maxY (vi),
y1=minY (vi),z0=maxZ (vi),z1=minZ (vi)
vi∈VS4
Finally, fitting submodule 31 is fitted face principal plane p0 on vertex set VS4, the plane area in bounding box Box0 is taken
Domain p1.Prenasale determines submodule 32 on plane domain in p1, and most preceding vertex is set to prenasale in plane p1 normal direction
Nosetip。
Wherein, face principal plane
P0:ax+by+cz+d=0, (a, b, c) are the normal direction of face dominant plane p0,
P1:ax+by+cz+d=0, x0≤x≤x1,y0≤y≤y1,z0≤z≤z1
(x0y0z0),(x1y1z1), be respectively Box0 two masters to angular vertex, p0 is that face master is fitted on vertex set VS4
Plane, p1 are the plane domain in bounding box Box0.
Prenasale definition:
Nosetip:Max (f (p)), wherein f (p)=ax+by+cz+d, p are a top in people on the face bounding box Box0
Point, (x, y, z) are its coordinate.
Since the distribution of nasal region curvature has polarization, top curvature value is larger, and marginal zone curvature is smaller, but because makes an uproar
The image of sound, compared to the biggish region of curvature, the lesser area distribution of curvature is more stable, by lesser using curvature value
Curvature detection face area and the effect for utilizing Gaussian curvature, average curvature and maximum principal curvatures realization to detect prenasale are very aobvious
It writes.
In above-mentioned multiple embodiments, which can be software unit, hardware cell or software and hardware combining list
Member.
Referring to Fig. 6, in the third embodiment of the present invention, the nose based on curvature distribution on a kind of three-dimensional face is provided
The detection method of point, the detection method include the following steps:
In step S601, computing module 10 constructs the grid surface of three-dimensional face, calculates the vertex on the network curved surface
Curvature and curvature distribution, and draw the corresponding curvature histogram of the curvature;
In step S602, filtering module 20 is pushed up according to the connectivity pair of the curvature distribution and the curvature histogram
Point carries out triple filtering processings, obtains filtered vertex set;
It in step S603, obtains module 30 and is fitted face principal plane on the filtered vertex set, and by the people
The vertex obtained in the normal direction in plane domain on face principal plane is positioned as the prenasale.
In this embodiment, first computing module 10 construct three-dimensional face grid surface, calculate on the network curved surface
Vertex curvature and curvature distribution, and draw the corresponding curvature histogram of the curvature;Filtering module 20 is then according to the song
Vertex described in connectivity pair of the rate distribution with the curvature histogram carries out triple filtering processings, obtains filtered vertex set;
It finally obtains module 30 and is fitted face principal plane on the filtered vertex set, and is flat by being obtained on the face principal plane
The vertex in normal direction in the region of face is positioned as the prenasale.
In the fourth embodiment of the present invention, the step S601 includes:
The image of the three-dimensional face of 11 pairs of submodule acquisitions of construction initializes, and constructs the net of the three-dimensional face images
Lattice curved surface;
First computational submodule 12 carries out denoising to the grid surface, and calculates the top on the network curved surface
Gaussian curvature, average curvature and two principal curvatures of point
Second computational submodule 13 is calculated corresponding using the Gaussian curvature or average curvature or two principal curvatures
Curvature distribution, and draw corresponding curvature histogram.
The curvature histogram are as follows:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the grade of histogram
Number, len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v
(ci), v is the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, put down
Equal one of curvature and two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viIt is
I vertex, i are the serial number on i-th of vertex.
The step 602 includes:
First filtering submodule 21 sets the upper threshold value th_t and lower threshold value th_b of the curvature histogram, according to the song
Rate distribution filters the curvature;
The vertex is divided by classification submodule 22 according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram
Three classes, wherein the vertex set less than the lower threshold value th_b is vertex set VS0, is in the lower threshold value th_b and upper threshold value th_
Vertex set is vertex set VS1 between t, and the vertex set greater than the upper threshold value th_t is vertex set VS2;
Second filtering submodule 23 is in the vertex set VS0 and vertex set VS2, according to the connection of the curvature histogram
Property search connection region, the connection component in the connection region is filtered, and only retains in the connection component most
Vertex set in the maximum connection component VS3 is labeled as VM by big connection component VS3;
Third filtering submodule 24 filters again in the vertex set VM, filters out the top from the vertex set VM
Vertex set VS3 in point set VS2, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and count
The bounding box of the vertex set VS4 is calculated, bounding box Box0 is labeled as.
The step S603 includes:
Fitting submodule 31 is fitted face principal plane p0 on the vertex set VS4, and takes in the bounding box Box0
Plane domain p1.
Prenasale determines submodule 32 on the plane domain in p1, and most preceding vertex is set to people in face p1 normal direction of making even
The prenasale of face.
In this embodiment, the image initial of the three-dimensional face of 11 pairs of submodule acquisitions, including construction grid song are constructed
Face, the denoising of the first computational submodule 12 and calculating vertex curvature, including Gaussian curvature, average curvature and two principal curvatures.Second
Computational submodule 13 calculates curvature distribution.
Wherein, the calculation formula of curvature:
Gaussian curvature:
Average curvature: H (vi)=(∑j∈N(i)(cotαij+cotβij)(vi-vj))/Avoronoi (2)
Wherein v in formula (1)iFor i-th of vertex on curved surface, N is vertex PiAdjoining number of triangles, θjIt is j-th
Adjacent triangular apex viAngle, AvoronoiIt is all of its neighbor triangle in vertex viThe sum of the Voronoi diagram shape area at place,
See Figure of description 3A.V in formula (2)i、vjI-th and j-th of vertex respectively on curved surface, N (i) are vertex viAdjoining
Vertex set, αijAnd βijTwo for side ij in two adjacent triangles are diagonal, AvoronoiIt is all of its neighbor triangle on vertex
viThe sum of Voronoi diagram shape area at place, see Figure of description 4B, AvoronoiCalculating see Figure of description 4A~figure
4C and Figure of description 5A~Fig. 5 G.Specifically, being distributed and calculating by average curvature since Fig. 5 A show original face
Afterwards, the average curvature distribution on Fig. 5 B three-dimensional face surface is obtained;Curvature distribution after bandpass filtering later, such as Fig. 5 C institute
Show;Then the maximum connection component processing of curvature distribution is carried out again, as shown in Figure 5 D;Then face principal plane and original are successively extracted
The principal plane of beginning face finally detects prenasale as shown in Fig. 5 E and Fig. 5 F, is nose institute as shown in arrow in Fig. 5 G
Show.
Principal curvatures:Its
Middle H (v) and G (v) are the average curvature and Gaussian curvature at vertex v respectively.
After completing curvature estimation, the first filtering submodule 21 sets the curvature histogram upper threshold value th_t and lower threshold value
Th_b filters curvature according to curvature distribution, and submodule 22 of classifying is divided into three classes according to th_t and th_b opposite vertexes, is less than th_b
Vertex set VS0, the vertex set VS1 and vertex set VS2 greater than th_t between th_b and th_t.Second filtering submodule 23 exists
Connection region is searched according to the connectivity of figure (what figure may I ask is) in vertex set VS0 and vertex set VS2, distich reduction of fractions to a common denominator amount carries out
Filtering, only retains maximum connection component, and vertex set is labeled as VM.Third filtering submodule 24 is filtered again in vertex set VM
Wave only retains the vertex set VS4 in VM in VS0 from the vertex set VS3 filtered out in VS2 in VM, and bounding box is labeled as
Box0。
Wherein, curvature histogram is defined as:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the grade of histogram
Number, len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v
(ci), v is the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, put down
Equal one of curvature and two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viIt is
I vertex, i are the serial number on i-th of vertex.
And vertex set is classified are as follows:
Vertex set VS0, VS0={ vi|Ci≤th_b},0≤i<N};
Vertex set VS1, VS1={ vi|th_b<Ci<th_t},0≤i<N};
Vertex set VS2, VS2={ vi|Ci≥th_t},0≤i<N};
Wherein CiFor vertex viThe curvature at place, N are vertex sum, and i is positive integer, and th_b, th_t are respectively curvature histogram
Lower threshold value and upper threshold value.
And it is connected to subset and largest connected subset are as follows:
VM=Max (len (VCi))
Wherein VCiFor i-th of connection subset, P (v on three-dimensional facek vh) it is vertex vk、vhBetween passage length, VM be most
Big connection subset, Max () are max function, and Len () calculates connected graph subset vertex quantity.
Maximum connection component filtering are as follows:
VS4=VM ∩ VS0;
Box0={ P0 P1},P0=(x0 y0 z0),P1=(x1 y1 z1);
x0=max X (vi),x1=min X (vi),y0=maxY (vi),
y1=minY (vi),z0=maxZ (vi),z1=minZ (vi),
vi∈VS4
Finally, fitting submodule 31 is fitted face principal plane p0 on vertex set VS4, the plane area in bounding box Box0 is taken
Domain p1.Prenasale determines submodule 32 on plane domain in p1, and most preceding vertex is set to prenasale in plane p1 normal direction
Nosetip。
Wherein, face principal plane
P0:ax+by+cz+d=0, (a, b, c) are the normal direction of face dominant plane p0,
P1:ax+by+cz+d=0, x0≤x≤x1,y0≤y≤y1,z0≤z≤z1
,(x0y0z0),(x1y1z1) be respectively Box0 two masters to angular vertex, p0 is that face master is fitted on vertex set VS4
Plane, p1 are the plane domain in bounding box Box0.
Prenasale definition:
Nosetip:Max (f (p)), wherein f (p)=ax+by+cz+d, p are a top in people on the face bounding box Box0
Point, (x, y, z) are its coordinate.
The detection method provided in above-mentioned multiple embodiments does not need to be aligned, directly it is advantageous that not needing to project
Capable calculating is tapped into, the time complexity of algorithm is low, O (nlogn), for four kinds of curvature, Gaussian curvature, average curvature and maximum
Principal curvatures effect is preferable.
On three-dimensional face curvature be distributed with its regularity, nasal region, Vitrea eye and mouth area are the concentration zones of minimax curvature
Domain, the present invention utilize the connectivity pair top of maximum curvature minimum distribution and three-dimensional face grid according to facial curvature distribution feature
Point is filtered three times, fitting face area principal plane, the method for positioning prenasale, and algorithm flow is as shown in Figure 7
Step S701, the initialization of three-dimensional face, including construction grid surface, denoising and calculating vertex curvature, including height
This curvature, average curvature and two principal curvatures.
Step S702 calculates curvature distribution.
Step S703 sets upper threshold value th_t and lower threshold value th_b, is filtered according to curvature distribution to curvature, according to th_t and
Th_bz opposite vertexes are divided into three classes, the vertex set VS0 less than th_b, vertex set VS1 and greater than th_t's between th_b and th_t
Vertex set VS2.
Step S704 searches connection area according to the connectivity of three-dimensional face grid in vertex set VS0 and vertex set VS2
Domain, distich reduction of fractions to a common denominator amount are filtered, and only retain maximum connection component, and vertex set is labeled as VM.
Step S705 is filtered again in vertex set VM, from the vertex set VS3 filtered out in VS2 in VM, is only retained in VM
Vertex set VS4 in VS0, bounding box are labeled as Box0.
Step S706 is fitted face principal plane p0 on vertex set VS4, takes the plane domain p1 in bounding box Box0.
Step S707, on plane domain in p1, most preceding vertex is set to prenasale Nosetip in plane p1 normal direction.
In this embodiment, it in order to verify the validity of the algorithm, has been carried out on the grid model of three-dimensional face corresponding
Experiment.The experimental calculation Gaussian curvature of all the points, average curvature and principal curvatures, according to the connectivity pair top of curvature distribution and figure
Point is filtered three times, and th_b and th_t is respectively 0.2 and 0.8 in experiment, and Gaussian curvature, average curvature also have similar
Effect.
In conclusion the grid surface of the invention by constructing three-dimensional face, calculates the vertex on the network curved surface
Curvature and curvature distribution, and draw the corresponding curvature histogram of the curvature;Then according to the curvature distribution and the song
Vertex described in the connectivity pair of rate histogram carries out triple filtering processings, obtains filtered vertex set;Finally, in the filtering
It is fitted face principal plane on vertex set afterwards, and the vertex in the normal direction obtained in plane domain on the face principal plane is determined
Position is the prenasale.Thus, it is possible to keep the detection of prenasale more easy, do not need specifically to project, does not need to be aligned, directly
Capable calculating is tapped into, the time complexity of algorithm is low, O (nlogn), for four kinds of curvature, Gaussian curvature, average curvature and maximum
Principal curvatures effect is preferable.
Certainly, the present invention can also have other various embodiments, without deviating from the spirit and substance of the present invention, ripe
It knows those skilled in the art and makes various corresponding changes and modifications, but these corresponding changes and change in accordance with the present invention
Shape all should fall within the scope of protection of the appended claims of the present invention.
Claims (6)
1. a kind of detection method of the prenasale on three-dimensional face based on curvature distribution, which is characterized in that the detection method packet
Include following steps:
A, the grid surface for constructing three-dimensional face, calculates the curvature and curvature distribution on the vertex on the network curved surface, and draw
The corresponding curvature histogram of the curvature;
B, the vertex according to the connectivity pair of the curvature distribution and the curvature histogram carries out triple filtering processings, obtains
Filtered vertex set;
C, it is fitted face principal plane on the filtered vertex set, and will be obtained in plane domain on the face principal plane
Normal direction on vertex be positioned as the prenasale;
Wherein the step A includes:
A1, the image of the three-dimensional face of acquisition is initialized, constructs the grid surface of the three-dimensional face images;
A2, denoising is carried out to the grid surface, and calculates the Gaussian curvature, average on vertex on the network curved surface
Curvature and two principal curvatures
A3, corresponding curvature distribution, and drafting pair are calculated using the Gaussian curvature or average curvature or two principal curvatures
The curvature histogram answered;
The step B includes:
The upper threshold value th_t and lower threshold value th_b of B1, the setting curvature histogram, according to the curvature distribution to the curvature
Filtering;
B2, the vertex is divided by three classes according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram, wherein be less than
The vertex set of the lower threshold value th_b is vertex set VS0, and vertex set is between the lower threshold value th_b and upper threshold value th_t
Vertex set VS1, the vertex set greater than the upper threshold value th_t are vertex set VS2;
B3, in the vertex set VS0 and vertex set VS2, according to the connectivity of the curvature histogram search connection region, it is right
Connection component in the connection region is filtered, and only retains the maximum connection component VS3 in the connection component, by institute
The vertex set in maximum connection component VS3 is stated labeled as VM;
B4, it is filtered again in the vertex set VM, from the vertex set filtered out in the vertex set VM in the vertex set VS2
VS3, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and calculate the vertex set VS4's
Bounding box is labeled as bounding box Box0.
2. detection method according to claim 1, which is characterized in that the step C includes:
C1, it is fitted face principal plane p0 on the vertex set VS4, and takes the plane domain p1 in the bounding box Box0;
C2, on the plane domain in p1, most preceding vertex is set to the prenasale of face in face p1 normal direction of making even.
3. detection method according to claim 1, which is characterized in that the curvature histogram are as follows:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the number of degrees of histogram,
Len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v(ci), v
For the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, average curvature and
One of two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viFor i-th of vertex,
I is the serial number on i-th of vertex.
4. a kind of detection system of the prenasale on three-dimensional face based on curvature distribution, which is characterized in that the detection system packet
It includes:
Computing module calculates the curvature and curvature on the vertex on the network curved surface for constructing the grid surface of three-dimensional face
Distribution, and draw the corresponding curvature histogram of the curvature;
Filtering module carries out triple filters for the vertex according to the connectivity pair of the curvature distribution and the curvature histogram
Wave processing, obtains filtered vertex set;
Module is obtained, for being fitted face principal plane on the filtered vertex set, and will be obtained on the face principal plane
The vertex in the normal direction in plane domain is taken to be positioned as the prenasale;
Wherein, the computing module includes:
Submodule is constructed, the image for the three-dimensional face to acquisition initializes, and constructs the net of the three-dimensional face images
Lattice curved surface;
First computational submodule for carrying out denoising to the grid surface, and calculates the top on the network curved surface
Gaussian curvature, average curvature and two principal curvatures of point
Second computational submodule, for calculating corresponding song using the Gaussian curvature or average curvature or two principal curvatures
Rate distribution, and draw corresponding curvature histogram;
The filtering module includes:
First filtering submodule, for setting the upper threshold value th_t and lower threshold value th_b of the curvature histogram, according to the song
Rate distribution filters the curvature;
Classification submodule, for the vertex to be divided into three according to the upper threshold value th_t and lower threshold value th_b of the curvature histogram
Class, wherein the vertex set less than the lower threshold value th_b is vertex set VS0, is in the lower threshold value th_b and upper threshold value th_t
Between vertex set be vertex set VS1, vertex set greater than the upper threshold value th_t is vertex set VS2;
Second filtering submodule is looked into the vertex set VS0 and vertex set VS2 according to the connectivity of the curvature histogram
Connection region is looked for, the connection component in the connection region is filtered, and only retains the maximum connection in the connection component
Vertex set in the maximum connection component VS3 is labeled as VM by reduction of fractions to a common denominator amount VS3;
Third filters submodule, filters again in the vertex set VM, filters out the vertex set from the vertex set VM
Vertex set VS3 in VS2, and only retain the vertex set VS4 in the vertex set VS0 in the vertex set VM, and calculate institute
The bounding box of vertex set VS4 is stated, bounding box Box0 is labeled as.
5. detection system according to claim 4, which is characterized in that the acquisition module includes:
It is fitted submodule, for being fitted face principal plane p0 on the vertex set VS4, and is taken flat in the bounding box Box0
Face region p1;
Prenasale determines submodule, in p1, most preceding vertex to be set to people in face p1 normal direction of making even on the plane domain
The prenasale of face.
6. detection system according to claim 4, which is characterized in that the curvature histogram are as follows:
FRE={ fre0,fre1,…,freM-1,
Wherein frejFor j-th of element in FRE, j ∈ { 0,1 ..., M-1 }, N are vertex sum, and M is the number of degrees of histogram,
Len=(Cmax-Cmin), maximum curvature Cmax=c | c=maxi∈v(ci), minimum curvature Cmin=c | c=mini∈v(ci), v
For the set { 0,1,2 ..., N-1 } of vertex serial number, ciFor vertex viThe curvature value at place, ciFor the Gaussian curvature, average curvature and
One of two principal curvatures, max () and min () are respectively the function for calculating maximum value and minimum value, viFor i-th of vertex,
I is the serial number on i-th of vertex.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610809148.9A CN106570450B (en) | 2016-09-07 | 2016-09-07 | The detection method and its system of prenasale on three-dimensional face based on curvature distribution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610809148.9A CN106570450B (en) | 2016-09-07 | 2016-09-07 | The detection method and its system of prenasale on three-dimensional face based on curvature distribution |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106570450A CN106570450A (en) | 2017-04-19 |
CN106570450B true CN106570450B (en) | 2019-10-18 |
Family
ID=58531789
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610809148.9A Active CN106570450B (en) | 2016-09-07 | 2016-09-07 | The detection method and its system of prenasale on three-dimensional face based on curvature distribution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106570450B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101131729A (en) * | 2007-09-25 | 2008-02-27 | 浙江大学 | Method for automatically extracting face area from three-dimensional scanning original data of head and shoulder |
CN102254343A (en) * | 2011-07-01 | 2011-11-23 | 浙江理工大学 | Convex hull and OBB (Oriented Bounding Box)-based three-dimensional grid model framework extracting method |
CN103236077A (en) * | 2013-04-15 | 2013-08-07 | 浙江理工大学 | Feature enhancement line drawing method for three-dimensional model |
CN103984920A (en) * | 2014-04-25 | 2014-08-13 | 同济大学 | Three-dimensional face identification method based on sparse representation and multiple feature points |
CN104463111A (en) * | 2014-11-21 | 2015-03-25 | 天津工业大学 | Three-dimensional face recognition method fused with multi-scale feature region curvatures |
CN105894561A (en) * | 2016-03-30 | 2016-08-24 | 中国科学院计算技术研究所 | Color mapping method and system on discrete curved surfaces based on curvature distribution |
-
2016
- 2016-09-07 CN CN201610809148.9A patent/CN106570450B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101131729A (en) * | 2007-09-25 | 2008-02-27 | 浙江大学 | Method for automatically extracting face area from three-dimensional scanning original data of head and shoulder |
CN102254343A (en) * | 2011-07-01 | 2011-11-23 | 浙江理工大学 | Convex hull and OBB (Oriented Bounding Box)-based three-dimensional grid model framework extracting method |
CN103236077A (en) * | 2013-04-15 | 2013-08-07 | 浙江理工大学 | Feature enhancement line drawing method for three-dimensional model |
CN103984920A (en) * | 2014-04-25 | 2014-08-13 | 同济大学 | Three-dimensional face identification method based on sparse representation and multiple feature points |
CN104463111A (en) * | 2014-11-21 | 2015-03-25 | 天津工业大学 | Three-dimensional face recognition method fused with multi-scale feature region curvatures |
CN105894561A (en) * | 2016-03-30 | 2016-08-24 | 中国科学院计算技术研究所 | Color mapping method and system on discrete curved surfaces based on curvature distribution |
Also Published As
Publication number | Publication date |
---|---|
CN106570450A (en) | 2017-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
CN106682598B (en) | Multi-pose face feature point detection method based on cascade regression | |
CN105447441B (en) | Face authentication method and device | |
Zhang et al. | Fast and robust occluded face detection in ATM surveillance | |
Cheng et al. | Exploiting effective facial patches for robust gender recognition | |
CN105894047B (en) | A kind of face classification system based on three-dimensional data | |
CN102592136B (en) | Three-dimensional human face recognition method based on intermediate frequency information in geometry image | |
CN108510516A (en) | A kind of the three-dimensional line segment extracting method and system of dispersion point cloud | |
CN109145717B (en) | Face recognition method for online learning | |
Li et al. | A comprehensive survey on 3D face recognition methods | |
Ming | Robust regional bounding spherical descriptor for 3D face recognition and emotion analysis | |
CN104143076B (en) | The matching process of face shape and system | |
Wang et al. | Head pose estimation with combined 2D SIFT and 3D HOG features | |
Rouhi et al. | A review on feature extraction techniques in face recognition | |
CN106778489A (en) | The method for building up and equipment of face 3D characteristic identity information banks | |
CN110516533A (en) | A kind of pedestrian based on depth measure discrimination method again | |
Cherian et al. | Bayesian nonparametric clustering for positive definite matrices | |
CN108446672A (en) | A kind of face alignment method based on the estimation of facial contours from thick to thin | |
Lin et al. | Local feature tensor based deep learning for 3d face recognition | |
CN103150575A (en) | Real-time three-dimensional unmarked human body gesture recognition method and system | |
Gilani et al. | Towards large-scale 3D face recognition | |
Alsawwaf et al. | In your face: person identification through ratios and distances between facial features | |
CN105550642B (en) | Gender identification method and system based on multiple dimensioned linear Differential Characteristics low-rank representation | |
Zheng et al. | Online subspace learning from gradient orientations for robust image alignment | |
CN110879985A (en) | Anti-noise data face recognition model training method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |