Fast shape retrieval based on shape contexts

Object shape is a powerful feature for content-based image retrieval. We propose a new shape contexts-based shape signature for shape retrieval. In terms of computing time, our method outperforms other approaches that calculate point correspondences based on shape contexts for shape retrieval. Consequently, our method is well suited for large databases. We demonstrate the retrieval performance of our approach on ETH-80 and KIMIA-216 data sets.