Component based shape retrieval using differential profiles

Abstract

In this paper, we describe the use of differential profiles, which are computed from 2D shapes smoothed with Gaussian functions, as the shape features for building a shape retrieval system. We build a global shape component dictionary from all the shape descriptors collected from shapes available in a database and then represent each shape as a probabilistic mixture of elements from such a dictionary. Finally, shape retrieval from a given database is simply done by comparing the mixing coefficients of the model of a query shape and those of known shapes. Our retrieval experiments are done on both object contour and line drawing collections and show promising results.

DOI: 10.1145/1460096.1460132

Extracted Key Phrases

8 Figures and Tables

Cite this paper

@inproceedings{Ding2008ComponentBS, title={Component based shape retrieval using differential profiles}, author={Lei Ding and Mikhail Belkin}, booktitle={Multimedia Information Retrieval}, year={2008} }