Shape Retrieval Based on Centroid Distance Ratio and Shape Context

Shape context is a classic shape retrieval method with translation invariance, but it has not scalar or rotational invariance, which limits its application. A new shape feature descriptor-centroid distance ratio (CdR) is proposed and an improved shape retrieval algorithm based on CdR and shape context is brought forth. First, the contour points are sampled in order to reduce computation and then, centroid distance values of the sample points are sorted in descending order. Ratio set of each pair of the list´s head and tail constructs the CdR feature for pre-matching. Final retrieval is achieved according to the fused feature. Experiments show that the proposed algorithm not only supports good translation, scalar, rotational invariance but also enhances evidently the retrieval efficiency and accuracy.