CN106971199A - A kind of automatic classification method of fabric three-dimensional draping shape - Google Patents

Abstract

The invention provides a kind of automatic classification method of fabric three-dimensional draping shape, it is characterised in that comprises the following steps：The pendency feature outlines of fabric three-dimensional draping shape are extracted, oval Fourier's description is carried out to the pendency feature outlines of three-dimensional drape form；Oval Fourier's description is carried out to the pendency feature outlines of the three-dimensional drape form of M sample in scanning fabric draping database, use is classified automatically without the hierarchy clustering method of given cluster number to three dimensional fabric overhang profiles line.The invention provides a kind of parametrization of general three dimensional fabric draping shape and automatic classification method, effectively clustering information can be extracted from three dimensional fabric draping shape, automatic classification is completed, so as to more comprehensive and accurate the relation analyzed between fabric draping form and physical property.

Description

A kind of automatic classification method of fabric three-dimensional draping shape

Technical field

The present invention relates to a kind of automatic classification method of fabric three-dimensional draping shape, for obtaining big by spatial digitizer After the threedimensional model for measuring true fabric draping, parameter processing is carried out to it and classified automatically, so as to serve fabric property Analysis, the reconstruction of virtual fabric and dress designing etc..

Background technology

The current test analysis to fabric static drape performance, which is concentrated mainly on, utilizes its plane projection to carry out image procossing. Because this method can not reflect the three-dimensional configuration feature of fabric completely, therefore with larger limitation.More for example have The fabric of similar suspended coefficient and node number but has entirely different physical property.

With the popularization of spatial digitizer so that the scanning to a large amount of true fabric drapings is possibly realized.At present, obtaining And establish three dimensional fabric pendency database after, have no idea realize classification.

The content of the invention

The purpose of the present invention is：Three-dimensional drape form to fabric carries out automatic Classification and Identification.

In order to achieve the above object, the technical scheme is that there is provided automatic point of a kind of fabric three-dimensional draping shape Class method, it is characterised in that comprise the following steps：

Step 1, the pendency feature outlines for extracting fabric three-dimensional draping shape, comprise the following steps：

Step 1.1, calculating fabric three-dimensional draping shape P_dIn X_DO_oY_DTwo-dimensional projection P in plane₁=f₁(P_d, Z_max), Z_maxIt is fabric three-dimensional draping shape P_dMaximum coordinate value among on Z axis, f₁(P_d, Z_max) it is by z ＜ Z_maxFabric three-dimensional pendency Form P_dThree dimensional point cloud project to X_DO_oY_DTo obtain two-dimensional projection P in plane₁Two-dimentional cloud data function；

Step 1.2, calculating two-dimensional projection P₁2-d contour C₀=f₂(P₁), f₂(P₁) it is to calculate two-dimensional projection P₁Two Tie up the function of contour line；

Calculate two-dimensional projection P₁Summit V₀=f₃(C₀), f₃(C₀) it is to calculate 2-d contour C₀Summit function；

Step 1.3, pass through summit V₀Coordinated indexing go out it in fabric three-dimensional draping shape P_dIn corresponding summit V₁, and Calculate V₁In min coordinates value Z on Z axis_min；

Step 1.4, the horizontal two-dimension contour line C for calculating three dimensional fabric pendency_i=f₂(f₁(P_d, Z_i)), i=1 ..., 9,Z₀For parameter set in advance；

Step 1.5, make 2-d contour C₀The Z coordinate of upper point is 200, horizontal two-dimension contour line C_iThe Z coordinate of upper point is Z_i, then 2-d contour C₀With horizontal two-dimension contour line C_iAs fabric three-dimensional draping shape P_dFeature outlines；

Step 2, the pendency feature outlines to three-dimensional drape form carry out oval Fourier and described, in the form of row vector (A₀, B₀, C₀, D₀, A₁, B₁, C₁, D₁... ..., A_n, B_n, C_n, D_n... ..., A_N, B_N, C_N, D_N) describe in contour curve, formula, n tables Show overtone order, N represents maximum overtone order, A_n、B_nRespectively pendency feature outlines are in projection plane X_DO_oY_DX-direction N-th harmonic oval coefficient, C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th The oval coefficient of harmonic wave；

Step 3, the pendency feature outlines to the three-dimensional drape form of M sample in scanning fabric draping database are carried out Oval Fourier's description, obtains matrix EFDs：

In formula, A_MN, B_MN, C_MN, D_MNRepresent m-th sample The oval coefficient of the n-th harmonic wave of the pendency feature outlines of this three-dimensional drape form, calculates E_EFDsCovariance matrix Characteristic value is adjacent with feature, E_EFDsFor oval Fourier descriptor, preceding k eigenvalue of maximum is constituted into projection matrix T_4N×k, then Obtain matrix EFDs k principal component PCs=E_EFDs×T_4N×k, k≤4N；

Step 4, the principal component obtained according to step 3, use is without the given hierarchy clustering method for clustering number to three-dimensional Fabric draping contour line is classified automatically.

Preferably, the A_n, B_n, C_n, D_nCalculation procedure be：

For pendency feature outlines, with it in X_DO_oY_DThe X of plane_DMinimum coordinate points on axle are starting point, using ellipse Circle Fourier methods are described as：

In formula, A₀、C₀Respectively in pendency feature outlines X, the y-coordinate of heart point, t are the arc length risen between point-to-point P of pendency feature outlines, and T is contour curve girth, and x (t) is t Functional relation between x coordinate, y (t) is the functional relation between t and y-coordinate；

A_n、B_nIt is to dangle feature outlines in projection plane X_DO_oY_DX-direction n-th harmonic oval coefficient, meter Calculation method is：

C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th harmonic oval system Count, computational methods are：

In formula, K is total for pendency feature outlines sampled point Number, Δ x_PFor pendency feature outlines on point P to point P+1 distance in X_DThe displacement of direction of principal axis, Δ y_PFor pendency feature contour The distance of point P to point P+1 on line is in Y_DThe displacement of direction of principal axis, Δ t_PFor pendency feature outlines on point P to point P+1 away from From that is,

A₀、C₀Respectively dangle x, the y-coordinate of feature outlines central point, and computational methods are：

In formula：

And ε₁=σ₁=0.

Preferably, the step 4 comprises the following steps：

Step 4.1, each sample is classified as a class, calculates the distance between each two class, that is, sample and sample it Between similarity, wherein the distance between r classes and s classes are d (r, s), then have：

In formula, T_rsIt is the Euclidean distance sum of all samples between any two in r classes and s classes, N_rAnd N_sNumber of samples in r classes and s classes respectively；

Step 4.2, nearest two classes between each class are found, they are classified as a class；

Step 4.3, recalculate similarity between this newly-generated class and each old class；

Step 4.4, repeat step 4.2 and step 4.3 are all classified as a class until all sample points, terminate.

, can be effective the invention provides a kind of parametrization of general three dimensional fabric draping shape and automatic classification method Ground extracts clustering information from three dimensional fabric draping shape, completes automatic classification, is knitted so as to more comprehensive and accurate analyze Relation between thing draping shape and physical property.

Brief description of the drawings

Fig. 1 (a) to Fig. 1 (c) is three dimensional fabric pendency indicatrix schematic diagram；

Fig. 2 is the cluster result that three dimensional fabric dangles；

Fig. 3 is the overhang profiles in class 1 near classification center；

Fig. 4 is the overhang profiles in class 2 near classification center；

Fig. 5 is the overhang profiles in class 3 near classification center；

Fig. 6 is the overhang profiles in class 4 near classification center；

Fig. 7 is the overhang profiles in class 5 near classification center.

Embodiment

With reference to specific embodiment, the present invention is expanded on further.It should be understood that these embodiments are merely to illustrate the present invention Rather than limitation the scope of the present invention.In addition, it is to be understood that after the content of the invention lectured has been read, people in the art Member can make various changes or modifications to the present invention, and these equivalent form of values equally fall within the application appended claims and limited Scope.

A kind of automatic classification method for fabric three-dimensional draping shape that the present invention is provided, comprises the following steps：

Step 1, the pendency feature outlines for extracting fabric three-dimensional draping shape, comprise the following steps：

Step 1.1, calculating fabric three-dimensional draping shape P_dIn X_DO_oY_DTwo-dimensional projection P in plane₁=f₁(P_d, Z_max), Z_maxIt is fabric three-dimensional draping shape P_dMaximum coordinate value among on Z axis, f₁(P_d, Z_max) it is by z ＜ Z_maxFabric three-dimensional pendency Form P_dThree dimensional point cloud project to X_DO_oY_DTo obtain two-dimensional projection P in plane₁Two-dimentional cloud data function, such as scheme Shown in 1 (a)；

Step 1.2, calculating two-dimensional projection P₁2-d contour C₀=f₂(P₁), f₂(P₁) it is to calculate two-dimensional projection P₁Two Tie up the function of contour line；

Calculate two-dimensional projection P₁Summit V₀=f₃(C₀), f₃(C₀) it is to calculate 2-d contour C₀Summit function；

Step 1.3, pass through summit V₀Coordinated indexing go out it in fabric three-dimensional draping shape P_dIn corresponding summit V₁, and Calculate V₁In min coordinates value Z on Z axis_min, such as shown in Fig. 1 (a)；

Step 1.4, the horizontal two-dimension contour line C for calculating three dimensional fabric pendency_i=f₂(f₁(P_d, Z_i)), i=1 ..., 9,Z₀For parameter set in advance；

Step 1.5, make 2-d contour C₀The Z coordinate of upper point is 200, horizontal two-dimension contour line C_iThe Z coordinate of upper point is Z_i, then 2-d contour C₀With horizontal two-dimension contour line C_iAs fabric three-dimensional draping shape P_dFeature outlines, such as Fig. 1 (b) And shown in Fig. 1 (c)；

Step 2, the pendency feature outlines to three-dimensional drape form carry out oval Fourier and described, in the form of row vector (A₀, B₀, C₀, D₀, A₁, B₁, C₁, D₁... ..., A_n, B_n, C_n, D_n... ..., A_N, B_N, C_N, D_N) describe in contour curve, formula, n tables Show overtone order, N represents maximum overtone order, A_n、B_nRespectively pendency feature outlines are in projection plane X_DO_oY_DX-direction N-th harmonic oval coefficient, C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th The oval coefficient of harmonic wave；

A_n, B_n, C_n, D_nCalculation procedure be：

For pendency feature outlines, with it in X_DO_oX_DThe X of plane_DMinimum coordinate points on axle are starting point, using ellipse Circle Fourier methods are described as：

In formula, A₀、C₀Respectively in pendency feature outlines X, the y-coordinate of heart point, t are the arc length risen between point-to-point P of pendency feature outlines, and T is contour curve girth, and x (t) is t Functional relation between x coordinate, y (t) is the functional relation between t and y-coordinate；

A_n、B_nIt is to dangle feature outlines in projection plane Y_DO_oY_DX-direction n-th harmonic oval coefficient, meter Calculation method is：

C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th harmonic oval system Count, computational methods are：

In formula, K is total for pendency feature outlines sampled point Number, Δ x_PFor pendency feature outlines on point P to point P+1 distance in X_DThe displacement of direction of principal axis, Δ y_PFor pendency feature contour The distance of point P to point P+1 on line is in Y_DThe displacement of direction of principal axis, Δ t_PFor pendency feature outlines on point P to point P+1 away from From that is,

A₀、C₀Respectively dangle x, the y-coordinate of feature outlines central point, and computational methods are：

In formula：

And ε₁=σ₁=0.

Step 3, the pendency feature outlines to the three-dimensional drape form of M sample in scanning fabric draping database are carried out Oval Fourier's description, obtains matrix EFDs：

In formula, A_MN, B_MN, C_MN, D_MNRepresent m-th sample The oval coefficient of the n-th harmonic wave of the pendency feature outlines of this three-dimensional drape form, calculates E_EFDsCovariance matrix Characteristic value is adjacent with feature, E_EFDsOval Fourier descriptor, projection matrix T is constituted by preceding k eigenvalue of maximum_4N×k, then To matrix EFDs k principal component PCs=E_EFDs×T_4N×k, k≤4N；

Step 4, the principal component obtained according to step 3, use is without the given hierarchy clustering method for clustering number to three-dimensional Fabric draping contour line is classified automatically.Step 4 comprises the following steps：

Step 4.1, each sample is classified as a class, calculates the distance between each two class, that is, sample and sample it Between similarity, wherein the distance between r classes and s classes are d (r, s), then have：

In formula, T_rsIt is the Euclidean distance sum of all samples between any two in r classes and s classes, N_rAnd N_sNumber of samples in r classes and s classes respectively；

Step 4.2, nearest two classes between each class are found, they are classified as a class；

Step 4.3, recalculate similarity between this newly-generated class and each old class；

Step 4.4, repeat step 4.2 and step 4.3 are all classified as a class until all sample points, terminate.

The present invention is illustrated with the sample of four kinds of unlike materials, and the basic performance of fabric sample is as shown in table 1.Survey Before examination, all static flattening of all samples is placed on 24 hours in the environment of relative humidity 65 ± 2%, 20 ± 2 DEG C of temperature.

The basic performance of the fabric sample of table 1

The woolen automatic classification results of draping shape of cotton, flax, silk, sample are as shown in Fig. 2 and table 2.

The class statistic result of the three dimensional fabric of table 2 pendency

Cluster result can be divided into obvious 5 major class as seen in Figure 2, all kinds of to account for sample total respectively 25.0%th, 31.8%, 11.4%, 27.3% and 4.5%.It can be seen from Table 2 that class 1 is main by low suspended coefficient and low node The pendency composition of number；Class 2 is mainly made up of the pendency of low suspended coefficient and high node number；Class 3 is main by high suspended coefficient Constituted with the pendency of high node number；Class 4 is mainly made up of the pendency of high suspended coefficient and low node number；Class 5 only has two Sample, for other samples, they have maximum suspended coefficient and minimum node number.In addition, we can also see Go out silk and principally fall into the 2nd class, and flax principally falls into the 4th class.Fig. 3 to 7 provide respectively it is all kinds of near classification center Overhang profiles.

Claims (3)

1. a kind of automatic classification method of fabric three-dimensional draping shape, it is characterised in that comprise the following steps：

Step 1, the pendency feature outlines for extracting fabric three-dimensional draping shape, comprise the following steps：

Step 1.1, calculating fabric three-dimensional draping shape P_dIn X_DO_oY_DTwo-dimensional projection P in plane₁=f₁(P_d, Z_max), Z_maxIt is to knit Thing three-dimensional drape form P_dMaximum coordinate value among on Z axis, f₁(P_d, Z_max) it is by z ＜ Z_maxFabric three-dimensional draping shape P_d's Three dimensional point cloud projects to X_DO_oY_DTo obtain two-dimensional projection P in plane₁Two-dimentional cloud data function；

Step 1.2, calculating two-dimensional projection P₁2-d contour C₀=f₂(P₁), f₂(P₁) it is to calculate two-dimensional projection P₁Two dimension wheel The function of profile；

Calculate two-dimensional projection P₁Summit V₀=f₃(C₀), f₃(C₀) it is to calculate 2-d contour C₀Summit function；

Step 1.3, pass through summit V₀Coordinated indexing go out it in fabric three-dimensional draping shape P_dIn corresponding summit V₁, and calculate Go out V₁In min coordinates value Z on Z axis_min；

Step 1.4, the horizontal two-dimension contour line C for calculating three dimensional fabric pendency_i=f₂(f₁(P_d, Z_i)), i=1 ..., 9,Z₀For parameter set in advance；

Step 1.5, make 2-d contour C₀The Z coordinate of upper point is 200, horizontal two-dimension contour line C_iThe Z coordinate of upper point is Z_i, then 2-d contour C₀With horizontal two-dimension contour line C_iAs fabric three-dimensional draping shape P_dFeature outlines；

Step 2, the pendency feature outlines to three-dimensional drape form carry out oval Fourier and described, (the A in the form of row vector₀, B₀, C₀, D₀, A₁, B₁, C₁, D₁... ..., A_n, B_n, C_n, D_n... ..., A_N, B_N, C_N, D_N) describe in contour curve, formula, n represents humorous Ripple number of times, N represents maximum overtone order, A_n、B_nRespectively pendency feature outlines are in projection plane X_DO_oY_DX-direction The oval coefficient of nth harmonic, C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th harmonic Oval coefficient；

Step 3, the pendency feature outlines to the three-dimensional drape form of M sample in scanning fabric draping database carry out oval Fourier describes, and obtains matrix EFDs：

In formula, A_MN, B_MN, C_MN, D_MNRepresent m-th sample The oval coefficient of the n-th harmonic wave of the pendency feature outlines of three-dimensional drape form, calculates E_EFDsCovariance matrix feature It is worth, E adjacent with feature_EFDsFor oval Fourier descriptor, preceding k eigenvalue of maximum is constituted into projection matrix T_4N×k, then obtain Matrix EFDs k principal component PCs=E_EFDs×T_4N×k, k≤4N；

Step 4, the principal component obtained according to step 3, use is without the given hierarchy clustering method for clustering number to three dimensional fabric Overhang profiles line is classified automatically.

2. a kind of automatic classification method of fabric three-dimensional draping shape as claimed in claim 1, it is characterised in that the A_n, B_n, C_n, D_nCalculation procedure be：

For pendency feature outlines, with it in X_DO_oY_DThe X of plane_DMinimum coordinate points on axle are starting point, using oval Fu In leaf method be described as：

$x\left(t\right)=A_0+\underset{n=1}{\overset{N}{\Σ}}\[A_{n}cos\left(\frac{2\mathrm{\π}nt}{T}\right)+B_{n}sin\left(\frac{2\mathrm{\π}nt}{T}\right)\];$

In formula, A₀、C₀Respectively dangle feature outlines central point X, y-coordinate, t is the arc length risen between point-to-point P of pendency feature outlines, and T is contour curve girth, and x (t) is that t and x are sat Functional relation between mark, y (t) is the functional relation between t and y-coordinate；

A_n、B_nIt is to dangle feature outlines in projection plane X_DO_oY_DX-direction n-th harmonic oval coefficient, calculating side Method is：

$A_n=\frac{T}{2{\mathrm{\π}}^2{n}^2}\underset{P=1}{\overset{K}{\Σ}}\frac{{\mathrm{\Δx}}_P}{{\mathrm{\Δt}}_P}(cos\frac{2{\mathrm{\πnt}}_P}{T}-cos\frac{2{\mathrm{\πnt}}_{P-1}}{T});$

$B_n=\frac{T}{2{\mathrm{\π}}^2{n}^2}\underset{P=1}{\overset{K}{\Σ}}\frac{{\mathrm{\Δx}}_P}{{\mathrm{\Δt}}_P}(\sin \frac{2{\mathrm{\πnt}}_P}{T}-\sin \frac{2{\mathrm{\πnt}}_{P-1}}{T});$

C_n、D_nRespectively pendency feature outlines are in projection plane X_DO_oY_DY direction n-th harmonic oval coefficient, meter Calculation method is：

$C_n=\frac{T}{2{\mathrm{\π}}^2{n}^2}\underset{P=1}{\overset{K}{\Σ}}\frac{{\mathrm{\Δy}}_P}{{\mathrm{\Δt}}_P}(cos\frac{2{\mathrm{\πnt}}_P}{T}-cos\frac{2{\mathrm{\πnt}}_{P-1}}{T});$

In formula, K is the sum of pendency feature outlines sampled point, Δx_PFor pendency feature outlines on point P to point P+1 distance in X_DThe displacement of direction of principal axis, Δ y_PFor pendency feature outlines On point P to point P+1 distance in Y_DThe displacement of direction of principal axis, Δ t_PFor pendency feature outlines on point P to point P+1 distance, I.e.

A₀、C₀Respectively dangle x, the y-coordinate of feature outlines central point, and computational methods are：

$A_0=\frac{1}{T}\underset{P=1}{\overset{K}{\Σ}}\[\frac{{\mathrm{\Δx}}_P}{2{\mathrm{\Δt}}_P}({t_P}^2-{t_{P-1}}^2)+\mathrm{\ϵ}(t_P-t_{P-1})\];$

In formula：

${\mathrm{\ϵ}}_P=\underset{j=1}{\overset{P-1}{\Σ}}{\mathrm{\Δx}}_j-\frac{{\mathrm{\Δx}}_P}{{\mathrm{\Δt}}_P}\underset{j=1}{\overset{P-1}{\Σ}}{\mathrm{\Δt}}_j;$

And ε₁=σ₁=0.

3. a kind of automatic classification method of fabric three-dimensional draping shape as claimed in claim 1, it is characterised in that the step 4 comprise the following steps：

Step 4.1, each sample is classified as a class, calculates the distance between each two class, that is, between sample and sample Similarity, wherein the distance between r classes and s classes are d (r, s), then has：

In formula, T_rsIt is r classes and the Euclidean distance sum of all samples between any two, N in s classes_rWith N_sNumber of samples in r classes and s classes respectively；

Step 4.2, nearest two classes between each class are found, they are classified as a class；

Step 4.3, recalculate similarity between this newly-generated class and each old class；

Step 4.4, repeat step 4.2 and step 4.3 are all classified as a class until all sample points, terminate.