Shape Feature Extraction
In this paper we present a method for automatic extraction of shape features, called crest lines. Shape features are important because they provide an alternative to describing an object, using its most important characteristics and reduce the amount of i
- PDF / 1,336,971 Bytes
- 13 Pages / 439 x 666 pts Page_size
- 25 Downloads / 236 Views
Abstract. In this paper we present a method for automatic extraction of shape features, called crest lines. Shape features are important because they provide an alternative to describing an object, using its most important characteristics and reduce the amount of information stored. The algorithm is comprised of a curvature approximation technique, crest point classification and a crest lines tracing algorithm.
1
Introduction
Three-Dimensional object comparison and matching are very important techniques with applications in robotics and automation, medicine, solid modelling, geometric modelling. It provides the ability to compare or match different objects and automatically identify their degree of similarity based on some distance criteria. Object matching attempts to find the transformation that best matches two given objects. In contrast, when doing object comparison a distance criterion is defined and used to compare different objects and return their distance. This method is usually used to rank objects based on their distance to a given object. Various methods have been developed in the past years that either use directly the object representation or transform the object to another domain before applying the matching algorithm. Hu [7] calculates all the possible rotations and translations of a scene object from a model object, represented as a connected graph of edges, and uses them to construct a three-dimensional Hough space. Johnson et al [8] use spin images of a model object to recognize a scene object. Shum et al [9] deform a sphere to the 3D closed surface, genus zero, and use curvature distribution, to define a similarity measure, for object comparison. Zhang and Hebert [10] perform surface matching after reducing the problem to 2D image-matching problem using harmonic maps. Declerck et al [11] extract shape features and use them to deform a source surface to a target surface. Most of these methods match objects and are limited within the domain of application. A notable exception is Shum et al that use a distance criterion for comparison but is applicable only to surfaces without holes. In this work, we extract shape features called crest lines that we will use in a later work to construct a feature based object comparison method. Crest lines are 3D lines on a surface that provide us with a satisfactory geometrical representation of important physical properties such as ridge lines and valleys G. Farin et al. (eds.), Hierarchical and Geometrical Methods in Scientific Visualization © Springer-Verlag Berlin Heidelberg 2003
270
G. Stylianou and G. Farin
in the case of aerial images [14], or anatomical features in the case of medical images [2,15]. They have been used, so far, for object registration [13], growth simulation [12] and automatic retrieval of anatomical structures [11]. We primarily focus on objects represented as triangulated meshes. With the advent of laser digitizers, we can easily acquire topologically different objects represented by triangulated meshes. Also, triangulations can represe
Data Loading...
