3D Video: From Capture to Diffusion

Chapter 18

3D Model Retrieval

18.1. Introduction

The use of three-dimensional (3D) models in multimedia data is expanding rapidly. The appearance of 3D scanners, the increasing power and simplicity of 3D creation tools and the arrival of 3D television (3DTV) all make it increasingly easy to create realistic and detailed 3D models. One question raised by the designers of virtual worlds concerns the best use of huge existing collections of 3D models, and efficient retrieval mechanisms for use with these collections.

The most immediately apparent solution to this problem is one already used to index images (by Google Images, for example): textual collection indexing, whereby each element in a collection is described using one or more keywords. This solution, although it appears simple, presents a number of drawbacks, including the time involved in tagging or labeling a collection and the subjective nature of keywords, which will inevitably be linked to the perception and culture of the tagging operator.

The most efficient method, and certainly the most intuitive method for final users of the retrieval system, is known as search by example. This method consists of characterizing 3D models using shape descriptors which are invariant to certain geometric transformations, notably translations, rotations and homothetic transformations. The user expresses a request in the form of an example of the object to find in a collection. A metric is then used to compare descriptors in the request ...