- Home
- Documents
*BLOWING BUBBLES FOR THE MULTI-SCALE ANALYSIS jarek/papers/ ¢ BLOWING BUBBLES FOR THE...*

prev

next

out of 22

View

0Download

0

Embed Size (px)

1

BLOWING BUBBLES FOR THE MULTI-SCALE ANALYSIS AND DECOMPOSITION OF TRIANGLE-MESHES

M. MORTARA �

, G. PATANÉ � , M. SPAGNUOLO

� , B. FALCIDIENO

� , J. ROSSIGNAC

�

ABSTRACT. Tools for the automatic decomposition of a surface into shape features will facilitate the editing, matching, texturing, morphing, compression, and simplification of 3D shapes. Different features, such as flats, limbs, tips, pits, and various blending shapes that transition between them may be characterized in terms of local curvature and other differ- ential properties of the surface or in terms of a global skeletal organization of the volume it encloses. Unfortunately, both solutions are extremely sensitive to small perturbations in the surface smoothness and to quantization effects when they operate on triangulated surfaces. Thus, we propose a multi-resolution approach, which not only estimates the curvature of a vertex over neighborhoods of variable size, but also takes into account the topology of the surface in that neighborhood. Our approach is based on blowing a spherical bubble at each vertex and studying how the intersection of that bubble with the surface evolves. We describe an efficient approach for computing these characteristics for a sampled set of bubble radii and for using them to identify features, based on easily formulated filters, that may capture the needs of a particular application.

keywords: Shape description, shape decomposition, multi-scale shape feature extrac- tion, shape indexing.

1. INTRODUCTION

Shape analysis and coding are challenging problems in Computer Vision and Graph- ics. An ideal shape description should be able to capture and compute the main features of a given shape and organize them into an abstract representation which can be used to automate processes such as matching, retrieval or comparison of shapes. We have tackled the problem in the context of 3D objects represented by triangular meshes, having in mind that a good shape description should be able to distinguish between global and local fea- tures and should be based on geometric properties of the shape which are invariant under rotation, translation and scaling [6]. To characterize a shape we have used the paradigm of blowing bubbles: a set of spheres of increasing radius ��� , ��� ��

��� �� �� is drawn, whose centers are at each vertex of the mesh, and whose radius represents the scale at which the shape is analyzed. The number of connected components of the intersection curve between each bubble and the surface gives a first qualitative characterization of the shape in a 3D neighborhood of each vertex. Then, the evolution of the ratio of the length of these com- ponents to the radius of the spheres is used to refine the classification and detect specific features such as sharp protrusions or wells, mounts or dips, blends or branching parts. For example, for a thin limb, that intersection will start simply connected and will rapidly split into two components. For a point on the tip of a limb, that intersection will usually simply remain connected, but the ratio of its length to the radius of the bubble will be decreasing.

1 � Istituto di Matematica Applicata e Tecnologie Informatiche, Consiglio Nazionale delle Ricerche, Genova, Italia. email: � falcidieno,michela,patane,spagnuolo � @ima.ge.cnr.it�

College of Computing and GVU Center, Georgia Institute of Technology, Atlanta. email: jarek@cc.gatech.edu

1

2 M. MORTARA � , G. PATANÉ � , M. SPAGNUOLO � , B. FALCIDIENO � , J. ROSSIGNAC �

For a point on a blend, that ratio will exceed ���

. An example of the resulting decomposi- tion is given in Figure 1.

FIGURE 1. Shape decomposition using blowing bubbles.

The achieved description provides an insight on the presence of features together with their morphological type, persistence at scale variation, amplitude and/or size. The decom- position is algorithmically simple, independent on the orientation of the object in space and equally distributed in all directions. The multi-scale approach and the chosen descriptors define a view-independent decomposition and reduce the influence of noise on the shape evaluation. The number of radii and the interval given by the minimum and maximum radius define the scope of the description for the input surface and the step between con- secutive scales. The choice of these parameters determines the final decomposition which is improved if a-priori knowledge about the size of the features to be extracted is available. In this paper we focus on the method adopted for the segmentation, while a possible ap- plication of the results can be found in [13], where a skeleton describing a shape from the point of view of its sharpest protrusions is presented.

The paper is organized as follows: in Section 2 previous work relevant for the described method is briefly reviewed. Basic concepts on differential geometry delineate, in Section 3, the theoretical background of the used geometric descriptors. The approach to shape classification is presented in Section 4. The algorithm description and details of the method are described in Section 5, while the mesh decomposition strategy is analyzed in Section 6. Finally, Section 7 includes critical considerations and remarks.

2. PREVIOUS WORK

An abstract description of a shape usually combines a set of primitives, that are relevant to the specific context, and is defined in terms of their type and intrinsic shape parameters.

As suggested in [12], methods for shape description can be classified into two broad cat- egories: those considering only the local properties of the boundary of the shape, and those measuring properties of the enclosed volume. Typically, boundary-based methods evalu- ate accurate and mathematically well-defined local characteristics, such as critical points or curvature. They may also identify specific loci on the surface, such as curvature extrema or ridges, but they generally lack in providing a global view of the shape. Furthermore, they typically work at a single resolution and thus do not organize features into a hierarchy of global and local details.

Conversely, interior-based methods, which assume that the surface is the boundary of a solid, generally provide descriptions which better highlight the global structure of the

BLOWING BUBBLES FOR THE MULTI-SCALE ANALYSIS AND DECOMPOSITION OF TRIANGLE-MESHES 3

shape. Skeletons, such as the medial axis or the Reeb graph [1, 16] belong to this class of descriptors. The great advantage of skeletons is that they provide an abstract representation by idealized lines that retain the connectivity of the original shape, thus reducing the com- plexity of the representation. Usually, each arc is associated with a portion of the original shape that corresponds to a feature. For example, in 2D the medial axis is constructed using the paradigm of the maximal enclosed disks, whose centers define a locus of points which describes, together with the associated radius, the width variation of the shape. The medial axis induces a decomposition of the shape into protrusion-like features, while concavities of the shape are not directly identified by the medial axis of the interior. Unfortunately, the medial axis of a 3D shape is not any longer a one-dimensional graph, but it is made of surface pieces as well. Moreover, the instability of the medial axis with respect to noise has prevented its use in many application areas. Approaches to construct and store the medial axis at different scales have been also proposed, which implicitly address the problem of noise reduction as well [16, 3, 15].

Another notable example of topology-driven skeleton is given by the Reeb graph [18, 1]. The Reeb graph is a topological structure which codes a given shape by storing the evolu- tion of criticalities of a mapping function defined on the boundary surface. In particular, when the height function with respect to a predefined direction is chosen, the Reeb graph describes the evolution of the contours obtained by intersecting the shape with constant planes. The decomposition induced by the Reeb graph corresponds to a segmentation of the solid into slices and the corresponding branches of the Reeb graph identify the con- nected components of the surface. The description obtained using a Reeb graph approach is suitable for matching purposes especially if the mapping function is chosen in order to provide invariance under affine transformations. Such orientation-independent approaches have been proposed in [10, 13]; however, they are computationally intensive and offer lit- tle control over the scale at which the shape is analyzed. We propose here an alternative and more efficient approach that gives us more flexibility to formulate the filters for shape analysis, and captures the more representative properties in a more detailed description.

3. THEORETICAL BACKGROUND

This section provides definitions and concepts [4, 8, 11, 14] useful for describing our approach. Let ��� ����� �� � ��� be a � � -parameterization of the surface

� � ��������� ���������� ���� � �"! � The classification of local properties of

� is traditionally based on the study of the mean

and the Gaussian curvature, which can be respectively defined as the average and the prod- uct of the maximum and minimum principal curvatures [11].

Let us consider the normal � to the surface � at a point # , and the normal sections of the surface around the normal vector, that is, the set of curves originated by intersecting the s