Semantic fitting and reconstruction

  • Authors:
  • Torsten Ullrich;Volker Settgast;Dieter W. Fellner

  • Affiliations:
  • Technical University Graz, Austria;Technical University Graz, Austria;Fraunhofer Institute for Computer Research and Technical University of Darmstadt, Germany

  • Venue:
  • Journal on Computing and Cultural Heritage (JOCCH)
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

The current methods to describe the shape of three-dimensional objects can be classified into two groups: methods following the composition of primitives approach and descriptions based on procedural shape representations. As a 3D acquisition device returns an agglomeration of elementary objects (e.g. a laser scanner returns points), the model acquisition pipeline always starts with a composition of primitives. Due to the semantic information carried with a generative description, a procedural model provides valuable metadata that make up the basis for digital library services: retrieval, indexing, and searching. An important challenge in computer graphics in the field of cultural heritage is to build a bridge between the generative and the explicit geometry description combining both worlds—the accuracy and systematics of generative models with the realism and the irregularity of real-world data. A first step towards a semantically enriched data description is a reconstruction algorithm based on decreasing exponential fitting. This approach is robust towards outliers and multiple dataset mixtures. It does not need a preceding segmentation and is able to fit a generative shape template to a point cloud identifying the parameters of a shape.