Shape Description and Retrieval in a Fused Scale Space

Abstract

In this work, a scale-space shape descriptor is proposed for shape retrieval, which is motivated by the multiscale mechanism of our human visual perception. First, morphological operations and the Gaussian smoothing are jointly used to produce a fused scale-space description of the input shape, which is able to handle strong noise, intra-class shape variation and irregular deformation simultaneously. Then, the height-function features of the shape are extracted across scales. Finally, shape retrieval is conducted by an integration of the retrieval results individually yielded at multiple scales. Experimental results on benchmark datasets validate the accuracy, efficiency and robustness of our proposed method.