Skip to Main content Skip to Navigation

Recherche par le contenu d'objets 3D

Abstract : This thesis deals with 3D shape similarity search. We focus on the main steps of the 3D shape matching process: normalization of 3D models, signature extraction from models, and similarity measure. The first part of the thesis concerns the normalization of 3D models, in particular the search for the optimal pose. We propose a new alignment method of 3D models based on the reflective symmetry and the local translational symmetry. We use the properties of the principal component analysis with respect to the planar reflective symmetry in order to select the eventual optimal alignment axes. The second part of the thesis is dedicated to the shape descriptors and the associated similarity measures. Firstly, we propose a new 3D descriptor, called 3D Gaussian descriptor, which is derived from the Gauss transform. Based on a partition of the enclosing 3D model space, this descriptor provides a local characterization of the boundary of the shape. Secondly, we study the multi-views based approaches that characterize the 3D model using their projection images. We introduce an augmented approach, named Enhanced Multi-views Approach, which can be applied in most of the multi-views descriptors. The relevance indices are defined and used in the similarity computation in order to normalize the contributions of the projections in the 3D-shape description. Finally, we propose a robust 3D shape indexing approach, called Depth Line Approach, which is based on the appearance of a set of depth-buffer images. To extract a compact signature, we introduce a sequencing method that transforms the depth lines into sequences. Retrieval is improved by using dynamic programming to compare sequence.
Complete list of metadata
Contributor : Ecole Télécom ParisTech Connect in order to contact the contributor
Submitted on : Tuesday, November 16, 2010 - 4:13:21 PM
Last modification on : Friday, July 31, 2020 - 10:44:05 AM
Long-term archiving on: : Thursday, February 17, 2011 - 2:31:19 AM


  • HAL Id : pastel-00005168, version 1



Mohamed Chaouch. Recherche par le contenu d'objets 3D. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2009. Français. ⟨pastel-00005168⟩



Record views


Files downloads