User-controllable mesh segmentation using shape harmonic signature
Yongwei Miao
a,b
, Jieqing Feng
a,
*
, Jinrong Wang
a,c
, Xiaogang Jin
a
a
State Key Laboratory of CAD & CG, Zhejiang University, Hangzhou 310027, China
b
College of Science, Zhejiang University of Technology, Hangzhou 310032, China
c
Information Science & Engineering School, Hangzhou Normal University, Hangzhou 310036, China
Received 14 April 2008; received in revised form 14 June 2008; accepted 16 June 2008
Abstract
Due to different shape modeling applications, partitioning a given complex 3D mesh model into some patches or meaningful subparts
is one of the fundamental problems in digital geometry processing. By using the high-dimensional mean-shift clustering scheme in shape
signature space, a new method is proposed which can generate user-specified segmentation results automatically for different applica-
tions. The shape signature is composed of mesh geometric attributes and its spectral harmonics. The latter one can reflect mesh frequency
spectrum information. The low frequency components are essential for semantics-oriented segmentation, while the high frequency com-
ponents are important for purely geometry-oriented segmentation. The effects of the proposed method are demonstrated by several
examples.
Ó 2008 National Natural Science Foundation of China and Chinese Academy of Sciences. Published by Elsevier Limited and Science in
China Press. All rights reserved.
Keywords: Mesh segmentation; Manifold harmonic analysis; Shape harmonic signature; Mean-shift clustering
1. Introduction
The problem of partitioning a polygonal mesh into
patches or meaningful subparts has been studied in digital
geometry process ing for different applications, such as tex-
ture mapping and texture atlas generation [1,2], surface
remeshing and mesh simplification [3,4], skeleton extrac-
tion [5], shape morphing and metamorphosis [6,7], shape
recognition and shape retrieval [8], and shape modeling [9].
Due to different applications, the mesh segmentation
methods can be classified into part-type ones (in a more
semantics-oriented manner) and patch-type ones (in a
purely geometric sense). The former aims at partitioning
the mesh into distinct semantic or meaningful parts accord-
ing to its shape features, e.g. a head and the legs of a horse,
without topological restriction for the segmented parts.
The latter one aims at partitioning the underlying mesh
into topological disk-like patches based on the mesh geom-
etry featu res, such as planarity, convexity and curvature.
According to the spectral analysis of a 3D model [10,11],
the high frequency signals contribute to the geometric
details, while the low frequency signals account for the
overall or global shape. In general, when the 3D spectral
information is considered, the mesh segmentation based
on low frequency information corresponds to the seman-
tics-oriented one, while the mesh segmentation based on
high frequency information corresponds to the geometry-
oriented one. The mesh segmentation method considered
in this paper focuses on both the aspects, and it is an
user-controllable scheme.
The proposed method is based on a high-dimensional
adaptive mean-shift clustering scheme in shape signature
space, where the shape signature is composed of mesh spec-
tral harmonics and its geometric attributes. Unlike the
traditional parametric segmentation methods, the pro-
posed one is a non-parametric mean-shift clustering
1002-0071/$ - see front matter Ó 2008 National Natural Science Foundation of China and Chinese Academy of Sciences. Published by Elsevier Limited
and Science in China Press. All rights reserved.
doi:10.1016/j.pnsc.2008.06.024
*
Corresponding author. Tel.: +86 571 88206681x506; fax: +86 571
88206680.
E-mail address: jqfeng@cad.zju.edu.cn (J. Feng).
www.elsevier.com/locate/pnsc
Available online at www.sciencedirect.com
Progress in Natural Science 19 (2009) 471–478