Statistiche



Three-Dimensional Shape Indexing and Retrieval Techniques

Pariticipating institutions

  • Department of Computer Science, University of Verona, Italy
  • Department of Computer and Information Science, University of Genova, Italy
  • Department of Mathematics, University of Cagliari, Italy
  • Department of Information Engineering, University of Padova, Italy
  • Department of Mathematics and Computer Science, University of Udine, Italy

Project Coordinator

Andrea Fusiello
Department of Computer Science, University of Verona

ImageHere.GIF

Aim of the project

Recent developments in techniques for modelling, digitizing and visualizing 3D shapes has led to an explosion in the number of available 3D models on the Internet and in domain-specific databases. This has led to the development of 3D shape retrieval systems that, given a query object, retrieve similar 3D objects.
Unlike text documents, 3D models are not easily retrieved. Attempting to find a 3D model using textual annotation and a conventional text-based search engine would not work in many cases. In contrast, content based 3D shape retrieval methods, that use shape properties of the 3D models to search for similar models, work better than text based methods.
Content-based shape retrieval consist of three main steps: descriptors extraction, indexing and matching. 3-SHIRT will address all these aspects, including also shape analysis (pre-processing) and overall evaluation. The issue of shape analysis consist of several techniques for the semantic representation of three-dimensional objects (models), like 1D skeletonization, features extraction, and segmentation.
One of major challenges is to elaborate a suitable canonical characterization (descriptor) of the entities to be indexed. Since the descriptor serves as a key for the search process, it decisively influences the performance of the search engine in terms of computational efficiency and relevance of the results. 3-SHIRT aims at devising new descriptors characterized by being i) invariant to more complex transformations ii) robust to occlusions iii) more closely related to human perception (and to human expectations) and iv) by including texture information as well as geometry. As for the matching phase, 3-SHIRT will focus on the design of matching methods with particular emphasis on the statistical learning approach for modelling the object’s class variability.
Content-based shape retrieval is a multidisciplinary field. The basic indexing and matching techniques exploit advanced and sophisticate methods and algorithms developed in the computer graphics and image processing communities. However, the overall problem is an Information Retrieval problem, and Information Retrieval provides both a framework for the whole research and specific techniques. This is particularly true for evaluation, which is an important facet of the project. In 3-SHIRT two evaluation stages are planned: a mid-project formative evaluation and a final evaluation at the end of the project. Formative evaluation will be a wide range evaluation: experiments based on different notions of relevance will be designed; user study and test collection with different metrics will be used. Final evaluation will concentrate on more specific issues emerged by the formative evaluation.

D.GT activities and tasks

Local coordinator / Contact

Guido Maria Cortelazzo

ImageHere.GIF

Creation of a collection of textured 3D models

For practical purposes the availability of a significant data-base of textured 3D objects is essential in order to validate the proposed descriptors of textured 3D models. No such a data-base is currently available for public usage. We envision to create our database from scratch using a set of models with no color information from freely available Web sites, or by direct acquisition of real objects.
One effort will be targeted at obtaining textured 3D models by adding either manually or, as much as possible, automatically various alternative textures to the downloaded models. Particular care must be paid in order to obtain textured models consistent with the selected similarity definition. For instance, corresponding parts must support similar texture if our similarity criterion was the one mentioned above. It can be noted that this approach has the advantage of a complete control on the similarity between colored 3D models and on the number of different categories, and it does not require manual annotations and categorization of models if the original shape models are already classified. A parallel effort will concern the direct acquisition of real textured objects by range cameras supporting color information.
Such 3D models will be added to the collection obtained by synthetically texturing the objects of existing collections of 3D models and consistently recorded within the existing classes of objects. The data set will be essential for the evaluations of the project resultst and will also be made available in the public and given its current uniqueness it will play a service for the community interested to content-based retrieval of textured 3D models.

ImageHere.GIF

Geometry and texture descriptors

This task investigates the extension of a non-textured 3D model descriptor (spin-images) in order to include texture information.
The use of spin-images for shape retrieval offers two advantages: firstly, it allows to exploit the similarities between object recognition and retrieval; secondly, the association of a set of 2D images to a 3D model makes it possible to benefit from the results achieved for image and video database retrieval. Starting from a suitable definition of similarity for textured 3D objects, several possible variations of spin-images will be defined and compared with respect to robustness, flexibility and retrieval efficiency.