![](bg1.png)
Skeleton-Based Shape Segmentation
Liudmila Domakhina
Moscow State University, Moscow, Russia
Abstract. Shape decomposition into meaningful parts (segmentation) problem
is considered in this paper. The shape is given either as a raster object on a ho-
mogeneous background or as a polygonal figure. A new shape decomposition
approach is called skeleton-based segmentation. The approach is based on con-
tinuous skeletons that provides an opportunity to construct visually proper seg-
mentations reflecting the shape structure. The proposed skeleton-based segmen-
tation method stands out against known methods because it is suitable to work
correctly with pairs of shapes. For a pair of shapes it is proposed to construct iso-
morphic skeleton-based segmentations which can be used on shape comparison
and morphing applications.
1 Introduction
Shape decomposition(segmentation) is a shape representation as a combination of com-
ponents (parts). The idea is to represent complex shapes in terms of simpler compo-
nents. If we have a decomposition of the shapes into parts, instead of matching the
shapes globally we could proceed by matching parts.
Shape segmentation is an important problem in many document processing applica-
tions (like organizingand querying an image database, recognition and computer-vision
problems, medical structure comparison etc.)
The goal of the present paper is to develop an effective segmentation method that
may be useful when working with pairs of shapes, especially in recognition and mor-
phing applications. Thus the segmentation should reflect the structure of the shape and
be stable to small shape fluctuations. Moreover, two similar shapes should have similar
segmentations. The segmentation method should be universal to work either with raster
objects or with polygonal figures.
The method presented in the paper is based on continuous skeletons. Continuous
skeleton reflects the shape structure but is not stable to boundary noise. The skeleton
could be considered as a graph. Skeletal graph could be transformed to become stable to
small shape variations. In case the shape is given as raster object on homogeneous back-
ground, the approximating poligonal figure and its skeleton could be constructed [1].
For a pair of shapes isomorphic skeletons and isomorphic skeleton-based segmentations
are constructed that allows solving such problems as morphing and shape comparison.
Domakhina L. (2009).
Skeleton-Based Shape Segmentation.
In Proceedings of the 2nd International Workshop on Image Mining Theory and Applications, pages 84-92
DOI: 10.5220/0001963500840092
Copyright
c
SciTePress