A Two - stage Matching Scheme for E ective and E

Recent content-based similar shape retrieval techniques are mostly based on either the global invariant statistics derived from the entire shape or the primitive features computed from a shape's local regions. Global features are easy to compute but do not accurately reeect a real shape, hence, have a limited discrimination ability. On the contrary, primitive local features are computationally intensive but ooer precise shape descriptions. We observed that the eeective-ness and eeciency of the similar shape retrieval can be enhanced by integrating global and local features embedded in the images. Using this approach, the retrieval is a two-stage process with the global features serve as an initial ltering mechanism to eliminate the signiicantly dissimilar images. The remaining images are then considered for the subsequent local feature based similarity matching. The two-stage retrieval mechanisms have been implemented in a metaserver. Extensive experiments have been conducted to demonstrate that the proposed designs ooer a viable and practical approach to similar shape retrieval.