Some Global Measures for Shape Retrieval

摘要

In this paper, we propose an efficient shape retrieval method. The idea is very simplistic, it is based on two global measures, the ellipse fitting and the minimum area rectangle. In this approach we don't need any information about the shape structure or its boundary form, as in most shape matching methods, we have only to compute the relativity between the surface of the shape and both of the minimum area rectangle encompassing it and its ellipse fitting. The proposed method is invariant to similarity transformations (translation, iso-tropic scaling and rotation). In addition, the matching gives satisfying results with minimal cost. The retrieval performance is illustrated using the MPEG-7 shape database.