Shape matching and retrieval via contour multi-scale decomposition

In this paper, we propose a new shape decomposition method based on visual perception for shape matching and retrieval which uses the protrusive corners under multiple scales. Our approach composes of a multi-scale shape decomposition procedure and a contour segments matching procedure. Firstly, morphological operations are used for more stable and visual perceptual corners to get the contour segmented under multiple scales. Secondly, a human perceptual custom and the sequence of segments are introduced and modeled in matching algorithm with segments described by SC descriptor. Finally, an ultimate similarity between two shapes is conducted by a fusion strategy based on multiple scales. Our method is evaluated on well-known shape databases including MPEG-7 and Kimia databases. The experimental retrieval results and the comparison with other methods clearly indicate the effectiveness of the proposed approach.