A Comparative Study of Three Region Shape Descriptors
Dengsheng Zhang, Guojun Lu · 2002
In Content Based Image Retrieval (CBIR), shape is one of the primary low level image features. Many shape representations have been proposed. However, most of them assume the knowledge of shape boundary information which is not available in general situations. Among them, region-based shape descriptors are not only applicable to generic shapes, but also robust to noise and distortions. In this paper we study and compare three region shape descriptors: Zernike moment descriptors (ZMD), grid descriptors (GD) and geometric moments descriptors (GMD). The strengths and limitations of these methods are analyzed and clarified. A Java frame retrieval framework is implemented to test the retrieval performance. The study and retrieval experiments on standard shape databases show that ZMD is the most suitable for shape retrieval in terms of computation complexity, compact representation, robustness, hierarchical coarse to fine representation and retrieval performance.