Shape matching using coarse descriptors

摘要

In this paper, we propose a new method for silhouettes matching. Silhouettes are described according to the textual language LWDOS (Larabi et al., 2003). Sometimes, although silhouettes appear identical, their LWDOS descriptors may be very different therefore we wrongly conclude that the two silhouettes are different. In order to resolve such problem, we try to deduce the same descriptor from two slightly different silhouettes descriptors to be compared. In this work, we use the part-based method published in Larabi et al. (2003) for object representation from its silhouettes. We use coarse descriptors in the matching process, this method is an efficient way for the indexing process for real and noisy images, but not applicable in this form to the recognition process, because different shapes can have same coarse descriptors.