B-spline surface reconstruction by inverse
subdivisions
Khoi NGUYEN-TAN
a
, Romain RAFFIN
b
, Marc DANIEL
c
LSIS, UMR CNRS 6168, 13288 Marseille cedex 09 – (France)
a
khoi.nguyen-tan@univmed.fr,
b
romain.raffin@lsis.org
c
marc.daniel@esil.univmed.fr
Cung LE
CRePA, PFIEV, DaNang University
DaNang, VietNam
k.spkt@ud.edu.vn
Abstract - This paper presents a method to reconstruct a B-spline
surface from a quadrangular mesh, using an inverse Catmull-
Clark subdivision. We want to minimize the surface contraction
due to the approximating subdivision scheme. We introduce
geometrical operations which minimize the impact of the
subdivision approximation and can be used in the parametric
surface reconstruction. The quality of the method is evaluated by
criteria of distances, curvatures or computing time on
experimental results.
Keywords : Reconstruction, B-spline surface, Catmull-Clark,
inverse subdivision, Data compression.
I. INTRODUCTION
In this paper we present a method to reconstruct a uniform
bicubic B-spline surface [7] from a quadrangular mesh using an
inverse Catmull-Clark subdivision. As the inverse subdivision
can be stopped after each step, different surfaces can be
obtained. The higher the number of steps, the better the data
reduction, but the larger the reconstruction error.
The two main contributions in this paper is an efficient
method to reconstruct a uniform bicubic B-spline surface using
an inverse Catmull-Clark subdivision scheme (UISS), and the
definition of criteria to evaluate the surface reconstruction
quality.
II. RELATED WORKS IN SURFACE RECONSTRUCTION
Reconstructing a parametric surface is a well-known
problem. [6] proposes an approximation algorithm using the
Catmull-Clark method for a NURBS surface. Based on this
work, [8] proposes a method to convert a Catmull-Clark
subdivision surface into a Bézier surface. In [9], the authors
apply the method in [6] to convert a surface into a NURBS
representation.
Our method is based on an inverse subdivision to
reconstruct a uniform bicubic B-spline surface from a
quadrangular initial mesh. As we consider the Catmull-Clark
subdivision scheme [1], it converges by definition towards a
uniform bicubic B-spline surface (called the limit surface). The
control polyhedron of this limit surface is the initial subdivision
mesh [2][5].
III. CATMULL-CLARK SUBDIVISION AND INVERSION
The main idea of the subdivision scheme is to define a
continuous surface (the limit surface) by successive steps on a
mesh. As the Catmull-Clark scheme is invertible, one can
restore recursively all the previous coarser meshes by using the
inverse subdivision scheme. We present here the inverse
Catmull-Clark subdivision scheme introduced in [4]. It
proposes the formulas to exactly reconstruct vertices of the
inverse mesh M
k
deduced from the vertices of mesh M
k+1
. At
each inverse subdivision, three distinguishable classes of points
are generated (see Figure 1), namely, the vertex points located
in the centre of four old face points (Figure 1a), the edge points
at the position corresponding of the old edge points (Figure 1b)
and the corner points in place corresponding of the face points
(Figure 1c).
1) Rules of inverse subdivision scheme
Considering the vertex points, the face points and the edge
points of quadrangular mesh M
k+1
, the relationship between the
vertices of quadrangular mesh M
k
and the corresponding
vertices of quadrangular mesh M
k+1
are deduced from the
previous equations.
The inverse vertex points of mesh M
k
are computed based
on formulas in [4] (Figure 1a):
( )
1 1 1 1 1
, 2 1,2 1 2 1,2 2 2 ,2 1 2 1,2 2 2,2 1
1 1 1 1
2 2,2 2 2 ,2 2 2 ,2 2 2,2
P 4P P +P P +P
P P P P
4
k k k k k k
ij i j i j i j i j i j
k k k k
i j i j i j i j
+ + + + +
- - - - - - - -
+ + + +
- - - -
= - +
+ + +
+
(1)
with i=1...m-2, j=1...n-2, where m=(h+3)%2, n=(l+3)%2.
The inverse edge points on the top border of mesh M
k
(except the corner points):
1 1 1
, 2 ,2 1 2 ,2 2 2 ,2 1,
P 4P P P P
k k k k k
ij i j i j i j i j
+ + +
- - +
= - - - (2)
with i=0, j=1..n-2; n=(l+3)%2. Other point edges of mesh
M
k
are computed similarly.
The inverse corner points: For four corner points P
00
, P
0,n-1
,
P
m-1,0
, P
m-1,n-1
of the mesh M
k
have a valence 2. They are
computed based on formula in [4]. Point P
00
is computed as
follows:
1
, 2 ,2 , 1 1, 1 1,
P 4P P P P
k k k k k
ij i j ij i j i j
+
+ + + +
= - - - with i=0, j=0 (3)
Other points are computed in the same way.
978-1-4244-4568-4/09/$25.00 ©2009 IEEE
1