Shape Distinction for 3D Object Retrieval
In recent years, there has been enormous growth in the number of 3D models and their availability to a wide segment of the population. Examples include the National Design Repository which stores 3D computer-aided design (CAD) models for tens of thousands of mechanical parts, the Protein Data Bank (PDB) that has atomic positions for tens of thousands of protein molecules, and the Princeton Shape Benchmark with thousands of everyday objects represented as polygonal surface models. With the availability of free interactive tools for creating 3D models and graphics cards for home computers, we can expect 3D data to become ever more widely available.
Given the availability of 3D data, searching for a 3D object in a large database is a core problem for numerous applications including object recognition and the reuse of expertly created data. This raises two key research problems: 1) How can we improve search techniques? and
2) How do we evaluate 3D search techniques?
The first contribution of this dissertation is an analysis technique to select the most important or distinctive regions of an object. Our approach identifies regions of a surface that have shape consistent with objects of the same type and different from objects of other types. By focusing a retrieval method on the most important regions of an object, we can improve retrieval performance in comparison to alternative feature point selection techniques. We investigate properties of shape distinction including techniques for calculating distinction, a method for visualizing differences in a database, and a prediction algorithm based on likelihoods of local shapes. We also demonstrate that shape distinction can be used in graphics applications such as mesh simplification and icon generation.
The second contribution is a new methodology to analyze shape retrieval methods with a common data set of classified 3D models and software tools called the Princeton Shape Benchmark (PSB). Based on experiments with several different retrieval methods, we find that no single method is best for all classifications of objects, and thus the main contribution of the PSB is a framework to evaluate retrieval methods.