Chapter 53
Estimation of Local Differential Properties

Table of Contents

This package allows the estimation of local differential quantities of a surface from a point sample, given either as a mesh or as point cloud.

Note that this package needs the third party libraries Lapack and Blas to be installed to compile the example code.

53.1   Introduction

53.1.1   Overview

Consider a sampled smooth surface, and assume we are given a collection of points P about a given sample p. We aim at estimating the differential properties up to any fixed order of the surface at point p from the point set P⁺ = P ∪ { p} - we denote N= | P⁺ | . More precisely, first order properties correspond to the normal or the tangent plane; second order properties provide the principal curvatures and directions, third order properties provide the directional derivatives of the principal curvatures along the curvature lines, etc. Most of the time, estimating first and second order differential quantities is sufficient. However, some applications involving shape analysis require estimating third and fourth order differential quantities. Many different estimators have been proposed in the vast literature of applied geometry [Pet01] (section 3, page 7), and all of them need to define a neighborhood around the point at which the estimation is computed. Our method relies on smooth differential geometry calculations, carried out on smooth objects fitted from the sample points. Datasets amenable to such a processing are naturally unstructured point clouds, as well as meshes - whose topological information may be discarded.

Estimating differential properties from discrete date always raises a philosophical issue. On one hand, estimating differential quantities subsumes a smooth surface does exist. In this spirit one wishes to recover its differential properties, so that any estimation method must come with an asymptotic convergence analysis of the results returned. For the method developed in this Cgal package, the interested will find such an analysis in [CP05a], (Theorem 3) - it should be stressed the error bounds proved therein are optimal.

On the other hand, any estimation method may be applied to arbitrarily data - surface unknown, surface piecewise smooth etc. In such a case, no analysis can be carried out, and it is up to the users to check the results match their needs.

Unlike most of the Cgal packages, this package uses approximation methods and is not intended to provide an exact canonical result in any sense. This is why internal computations are performed with a number type possibly different from that of the input data, even if for convenience the results are returned with this original number type. A reasonable choice for this internal number type is for example the double type.

53.1.2   Smooth Surfaces, d-Jets and the Monge Form

To present the method, we shall need the following notions. Consider a smooth surface. About one of its points, consider a coordinate system whose z-axis does not belong to the tangent space. In such a frame, the surface can locally be written as the graph of a bivariate function. Letting h.o.t. stand for higher order terms, one has :

Recall that an umbilical point of a surface - or umbilic for short, is a point where both principal curvatures are identical. At any point of the surface which is not an umbilic, principal directions d₁, d₂ are well defined, and these (non oriented) directions together with the normal vector n define two direct orthonormal frames. If v₁ is a unit vector of direction d₁, there exists a unique unit vector v₂ so that (v₁,v₂,n) is direct; and the other possible frame is (-v₁,-v₂,n). Both these coordinate systems are known as the Monge coordinate systems. In both these systems, the surface is said to be given in the Monge form and its jet has the following canonical form :

The coefficients k₁, k₂ are the principal curvatures, b₀,b₃ are the directional derivatives of k₁,k₂ along their respective curvature line, while b₁,b₂ are the directional derivatives of k₁,k₂ along the other curvature lines.

The Monge coordinate system can be computed from any d-jet (d ≥ 2), and so are the Monge coefficients. These informations characterize the local geometry of the surface in a canonical way, and are the quantities returned by our algorithm.

53.1.3   Algorithm

1. We perform a Principal Component Analysis (PCA) on P⁺. This analysis outputs three orthonormal eigenvectors and the associated eigenvalues. The fitting basis consists of these three vectors so that the vector associated to the smallest eigenvalue is the last vector of the basis. (Indeed, if the surface is well sampled, one expects the PCA to provide one small and two large eigenvalues, the eigenvector associated to the small one approximating the normal vector.)
2. We perform a change of coordinates to move the samples into the coordinate system of the fitting basis and with origin the point p at which the estimation is sought. We then resort to polynomial fitting, so as to either interpolate or approximate the d-jet of the surface in this coordinate system. This bivariate polynomial approximation reduces to linear algebra operations.
3. From the fitted d-jet, we compute the Monge basis (d₁,d₂,n).
4. Finally, we compute the Monge coefficients : k_i, b_i, c_i.

Further details can be found in section 53.4 and in [CP05a] (section 6).

53.1.4   Degenerate Cases

As usual, the fitting procedure may run into (almost) degenerate cases:

• Due to poor sampling, the PCA used to determine a rough normal vector may not be good. The nearer this direction to the tangent plane the worse the estimation.

• As observed in [CP05a] (section 3.1), the interpolating problem is not poised if the points project, into the fitting frame, onto an algebraic curve of degree d. More generally, the problem is ill poised if the condition number is too large.

53.2   Software Design

53.2.1   Options and Interface Specifications

The fitting strategy performed by the class Monge_via_jet_fitting requires the following parameters:

• the degree d of the fitted polynomial (d ≥ 1),
• the degree d' of the Monge coefficients sought, with 1 ≤ d' ≤ min (d,4) ,
• a range of N input points on the surface, with the precondition that N ≥ N_d = (d+1)(d+2)/2. Note that if N=N_d, interpolation is performed; and if N >N_d, approximation is used.

53.2.2   Output

As explained in Section 53.1, the output consists of a coordinate system, the Monge basis, together with the Monge coefficients which are stored in the Monge_form class. In addition, more information on the computational issues are stored in the Monge_via_jet_fitting class.

The Monge_form class provides the following information.

• Origin. This is the point on the fitted polynomial surface where the differential quantities have been computed. In the approximation case, it differs from the input point p : it is the projection of p onto the fitted surface following the z-direction of the fitting basis.

• Monge Basis. The Monge basis (d₁,d₂,n) is orthonormal direct, and the maximal, minimal curvatures are defined wrt this basis. If the user has a predefined normal n₀ (e.g. the sample points come from an oriented mesh) then if n₀ . n >0 then max-min is correct; if not, i.e. n₀ . n <0, the user should switch to the orthonormal direct basis (d₁',d₂',n')=(d₂,d₁,-n) with the maximal curvature k₁'=-k₂ and the minimal curvature k₂'=-k₁. If n₀ . n =0 or is small, the orientation of the surface is clearly ill-defined, and the user may proof-check the samples used to comply with its predefined normal.

• Monge Coefficients. The coefficient of the Monge form is (k₁, k₂ ( ≤ k₁), b₀, b₁, b₂, b₃, c₀, c₁, c₂, c₃, c₄) for d' = 4.

In addition, the class Monge_via_jet_fitting stores

• the condition number of the fitting system,
• the coordinate system of the PCA (in which the fitting is performed).

53.2.3   Template Parameters

Template parameter DataKernel

This concept provides the types for the input sample points, together with 3d vectors and a number type. It is used as template for the classe Monge_via_jet_fitting<DataKernel, LocalKernel = Cartesian<double>, SvdTraits = lapack_svd> . Typically, one can use CGAL::Cartesian<double>.

Template parameter LocalKernel

Input points of type DataKernel::Point_3 are converted to LocalKernel::Point_3. For output of the Monge_form class, these types are converted back to Data_Kernel ones. Typically, one can use CGAL::Cartesian<double> which is the default.

Template parameter SvdTraits

This concept provides the number, vector and matrix types for algebra operations required by the fitting method in Monge_via_jet_fitting<DataKernel, LocalKernel = Cartesian<double>, SvdTraits = lapack_svd> . The main method is a linear solver using a singular value decomposition.

Compatibility requirements

To solve the fitting problem, the sample points are first converted from the DataKernel to the LocalKernel (this is done using the CGAL::Cartesian_converter). Then change of coordinate systems and linear algebra operations are performed with this kernel. This implies that the number types LocalKernel::FT and SvdTraits::FT must be identical. Second the Monge basis and coefficients, computed with the LocalKernel, are converted back to the DataKernel (this is done using the CGAL::Cartesian_converter and the CGAL::NT_converter).

53.3   Examples

53.3.1   Single Estimation about a Point of a Point Cloud

The first example illustrates the computation of the local differential quantities from a set of points given in a text file as input. The first point of the list is the one at which the computation is performed. The user has to specify a file for the input points and the degrees d and d'.

File: examples/Jet_fitting_3/Single_estimation.cpp

#include <CGAL/Cartesian.h>


#include <fstream>
#include <vector>

#include <CGAL/Monge_via_jet_fitting.h>
typedef double                   DFT;
typedef CGAL::Cartesian<DFT>     Data_Kernel;
typedef Data_Kernel::Point_3     DPoint;
typedef CGAL::Monge_via_jet_fitting<Data_Kernel> My_Monge_via_jet_fitting;
typedef My_Monge_via_jet_fitting::Monge_form     My_Monge_form;

int main(int argc, char *argv[])
{
  size_t d_fitting = 4;
  size_t d_monge = 4;
  const char* name_file_in = "data/in_points_d4.txt";
  //check command line
  if (argc<4)
    {
      std::cout << " Usage : " << argv[0]
                << " <inputPoints.txt> <d_fitting> <d_monge>" << std::endl
		<< "test with default arguments" << std::endl;
    }
  else {
    name_file_in = argv[1];
    d_fitting = std::atoi(argv[2]);
    d_monge = std::atoi(argv[3]);
  }

  //open the input file
  std::ifstream inFile(name_file_in);
  if ( !inFile )
    {
      std::cerr << "cannot open file for input\n";
      exit(-1);
    }
  
  //initalize the in_points container
  double x, y, z;
  std::vector<DPoint> in_points;
  while (inFile >> x) {
    inFile >> y >> z;
    DPoint p(x,y,z);
    in_points.push_back(p);
  }
  inFile.close();
  // fct parameters

  My_Monge_form monge_form;
  My_Monge_via_jet_fitting monge_fit;
  monge_form = monge_fit(in_points.begin(), in_points.end(), d_fitting, d_monge);

  //OUTPUT on std::cout
  CGAL::set_pretty_mode(std::cout);
  std::cout << "vertex : " << in_points[0] << std::endl
	    << "number of points used : " << in_points.size() << std::endl
	    << monge_form;
  std::cout  << "condition_number : " << monge_fit.condition_number() << std::endl
	     << "pca_eigen_vals and associated pca_eigen_vecs :"  << std::endl;
  for (int i=0; i<3; i++)
    std::cout << monge_fit.pca_basis(i).first << std::endl
	      << monge_fit.pca_basis(i).second  << std::endl;
  return 0;
}

53.3.2   On a Mesh

• a human readable text file featuring the Monge_form and numerical informations on the computation : condition number and the PCA basis;
• another text file that records raw data (better for a vizualization post-processing).

Figs. 53.1 and 53.2 provide illustrations of principal directions of curvature.

53.4   Mathematical and Algorithmic Details

In this Section, we detail the mathematics involved, in order to justify the design choices made. To begin with, observe the fitting problem involves three relevant direct orthonormal basis: the world-basis (w_x,w_y,w_z), the fitting-basis (f_x,f_y,f_z), the Monge basis (d₁,d₂,n).

Figure 53.3:  The three bases involved in the estimation.

53.4.1   Computing a Basis for the Fitting

Input : samples
Output : fitting-basis

Performing a PCA requires diagonalizing a symmetric matrix. This analysis gives an orthonormal basis whose z-axis is provided by the eigenvector associated to the smallest eigenvalue.¹ Note one may have to swap the orientation of a vector to get a direct basis.

Let us denote P_{W → F} the matrix that changes coordinates from the world-basis (w_x,w_y,w_z) to the fitting-basis (f_x,f_y,f_z). The rows of P_{W → F} are the coordinates of the vectors (f_x,f_y,f_z) in the world-basis. This matrix represents a orthogonal transformation hence its inverse is its transpose. To obtain the coordinates of a point in the fitting-basis from the coordinates in the world-basis, one has to multiply by P_{W → F}.

As mentioned above, the eigenvalues are returned, from which the sampling quality can be assessed. For a good sampling, the eigenvector associated to the smallest eigenvalue should roughly give the normal direction.

53.4.2   Solving the Interpolation / Approximation Problem

Input : samples, fitting-basis
Output : coefficients A_i,j of the bivariate fitted polynomial in the fitting-basis

Computations are done in the fitting-basis and the origin is the point p. First, one has to transform coordinates of sample points with a translation (-p) and multiplication by P_{W → F}.

The fitting process consists of finding the coefficients A_i,j of the degree d polynomial

JA,d=

d ∑ k=0 ( k ∑ i=0 Ak-i,ixk-iyi i!(k-i)! )

.

Denote p_i=(x_i,y_i,z_i), i=1, … , N the coordinates of the sample points of P⁺. For interpolation the linear equations to solve are A(x_i,y_i)=z_i i=1, … ,N, and for approximation one has to minimize ∑_i=1^N (A(x_i,y_i)-z_i)². The linear algebra formulation of the problem is given by

A = (A0,0, A1,0,A0,1, … , A0,d)T Z= (z1, z2, … , zN)T M= (1,xi, yi, xi2 2 , … , xiyid-1 (d-1)! , yid d! )i=1,...,N

The equations for interpolation become MA=Z. For approximation, the system MA=Z is solved in the least square sense, i.e. one seeks the vector A such that A = arg min _A ||MA-Z||₂.

In any case, there is a preconditioning of the matrix M so as to improve the condition number. Assuming the {x_i}, {y_i} are of order h, the pre-conditioning consists of performing a column scaling by dividing each monomial x_i^ky_i^l by h^k+l - refer to Eq. (53.4.2). Practically, the parameter h is chosen as the mean value of the {x_i} and {y_i}. In other words, the new system is M'Y=(MD^-1)(DA)=Z with D the diagonal matrix D=(1,h,h,h², … ,h^d,h^d), so that the solution A of the original system is A=D^-1Y.

There is always a single solution since for under constrained systems we also minimize ||A||₂. The method uses a singular value decomposition of the N × N_d matrix M= U S V^T, where U is a N × N orthogonal matrix, V is a N_d × N_d orthogonal matrix and S is a N × N_d matrix with the singular values on its diagonal. Denote r the rank of M, we can decompose S= (

Dr	0r, Nd-r
0N-r, r	0N-r, Nd-r

). The number r, which is the number of non zero singular values, is strictly lower than N_d if the system is under constrained. In any case, the unique solution which minimize ||A||₂ is given by :

A= V (

Dr-1 0Nd-r, r 0r, N-r 0Nd-r, N-r

) UTZ.

One can provide the condition number of the matrix M (after preconditioning) which is the ratio of the maximal and the minimal singular values. It is infinite if the system is under constrained, that is the smallest singular value is zero.

Implementation details. We assume a solve function is provided by the traits SvdTraits. This function solves the system MX=B (in the least square sense if M is not square) using a Singular Value Decomposition and gives the condition number of M.

Remark: as an alternative, other methods may be used to solve the system. A QR decomposition can be substituted to the SVD. One can also use the normal equation M^TMX=MTB and apply methods for square systems such as LU, QR or Cholesky since M^TM is symmetric definite positive when M has full rank. The advantages of the SVD is that it works directly on the rectangular system and gives the condition number of the system. For more on these alternatives, see [GvL83] (Chap. 5).

53.4.3   Principal Curvature / Directions

Input : coefficients of the fit A_i,j, fitting-basis
Output : Monge basis wrt fitting-basis and world-basis

In the fitting basis, we have determined a height function expressed by Eq. (53.4.2). Computations are done in the fitting-basis. The partial derivatives, evaluated at (x,y)=(0,0), of the fitted polynomial J_A,d(x,y) are A_i,j=(∂^i+jJ_A,d)/(∂ⁱx ∂^jy) Expanding Eq. (53.4.2) yields:

JA,d(x,y) = A0,0+A1,0x+A0,1y+ 1 2 (A2,0x2+2A1,1xy+A0,2y2) + 1 6 (A3,0x3+3A2,1x2y+ … )+ …

• The origin, that is the point of the fitted surface where the estimation is performed, is (0,0,A_0,0).
• The normal is n=(-A_1,0,-A_0,1,1)/√A_1,0²+A_0,1²+1.
• Curvature related properties are retrieved resorting to standard differential calculus [dC76] (Chap. 3). More precisely, the Weingarten operator W=-I^-1II is first computed in the basis of the tangent plane { (1,0,A_1,0), (0,1,A_0,1) }. We compute an orthonormal basis of the tangent plane using the Gram-Schmidt algorithm, and then we compute Weingarten in this basis (applying a change of basis with the matrix W'=P^-1WP). In this orthonormal basis, the matrix of the Weingarten map is symmetric and we diagonalize it. One finally gets the principal curvatures which are the eigenvalues of W, and the associated principal directions. This gives an orthonormal direct basis (d₁,d₂,n). Let us denote P_{F → M} the matrix to change coordinates from the fitting-basis to the Monge basis. Its rows are the coordinates of the vectors (d₁,d₂,n) in the fitting-basis. It is an orthogonal matrix P_{F → M}^-1=P_{F → M}^T. The Monge basis expressed in the world-basis is obtained by multiplying the coordinates of (d₁,d₂,n) in the fitting-basis by P_{W → F}^-1, (the same holds for the origin point which has in addition to be translated by p, i.e. the coordinates of the origin point are P_{W → F}^-1 (0,0,A_0,0) +p.

53.4.4   Computing Higher Order Monge Coefficients

Input : coefficients of the fit, Monge basis wrt fitting-basis (P_{F → M})
Output : third and fourth order coefficients of Monge

We use explicit formula. The implicit equation of the fitted polynomial surface in the fitting-basis with origin the point (0,0,A_0,0) is Q=0 with

Q=-w-A0,0 +

∑ i,j Ai,juivj i!j!

.

The equation in the Monge basis is obtained by substituting (u,v,w) by P^T_{F → M}(x,y,z). Denote f(x,y,z)=0 this implicit equation. By definition of the Monge basis, we have locally (at (0,0,0))

f(x,y,z)=0 ⇔ z=g(x,y)

and the Taylor expansion of g at (0,0) are the Monge coefficients sought. Let us denote the partial derivatives evaluated at the origin of f and g by f_i,j,k=(∂^i+j+kf)/(∂ⁱx ∂^jy ∂^kz) and g_i,j=(∂^i+jg)/(∂ⁱx ∂^jy). One has f_1,0,0=f_0,1,0=f_1,1,0=0, g_0,0=g_1,0=g_0,1=g_1,1=0 and g_2,0=k₁, g_0,2=k₂. The partial derivative of order n of f depends on the matrix P_{F → M} and the partial derivatives of order at most n of J_A,d. The third and fourth order coefficients of are computed with the implicit function theorem. For instance :

b0=g3,0=- f3,0,0 f0,0,1 -3 f1,0,1 f2,0,0 f0,0,1 2 b1=g2,1=- - f0,1,1 f2,0,0 + f2,1,0 f0,0,1 f0,0,1 2 ....

Footnotes