Implementation of Morphological Operators for Surface Segmentation - Archive ouverte HAL Accéder directement au contenu
Rapport Année : 2007

Implementation of Morphological Operators for Surface Segmentation

Résumé

A content-based retrieval system needs feature vectors for database indexation. We adopt the surface segmentation approach to obtain several features for a 3D object which can be used to retrieve objects in a database from a partial request composed from a set of regions. To achieve the segmentation of the surface into several regions we apply the watershed transform on a curvature map computed on the 3D surface mesh. The watershed applied directly on the original curvature map produces an over-segmentation of the object surface. Thus, we previously filter the original curvature map by using the component tree. After this filtering, the watershed transform is computed on the filtered curvature map and we obtain the desired number of regions. Then we proceed by computing some features for each region obtained, which will serve as feature vectors for a content-based search and retrieval system. The techniques we apply on the surface of 3D objects have been presented for image applications. Consider a 3D triangular mesh (a set of points, triangles and sides of triangles). We build an edge weighted graph to represent the mesh. Weights on edges of the graph correspond to the curvature map computed on the mesh. The component tree is a structure to order the connected components of a weighted graph. We implement an original algorithm to build the component tree of an edge weighted graph based on the one presented by Najman and Couprie [14]. This structure allows to reduce the number of minima of the original map on edges of the graph. We implement the algorithm proposed by Cousty [5] to compute the watershed transform on an edge weighted graph. By using the watershed transform we obtain a number of regions of the map which equals the number of minima of the input map. Once we obtain the partition of the mesh into several regions we compute features of each region. These features consist in histograms for each region considering three differ- ent approaches: Extended Gaussian Images (EGI) [8]; a cords histogram (considering the cords lenght and a principal angle); a curvature histogram (considering principal curva- tures). These histograms form feature vectors for each region which will help the database indexation and classification.
Fichier principal
Vignette du fichier
surface_m_alcoverro.pdf (1.83 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-00749115 , version 1 (06-11-2012)

Identifiants

  • HAL Id : hal-00749115 , version 1

Citer

Marcel Alcoverro. Implementation of Morphological Operators for Surface Segmentation. 2007. ⟨hal-00749115⟩

Collections

LARA
45 Consultations
45 Téléchargements

Partager

Gmail Facebook X LinkedIn More