2-D Shape Matching Using Asymmetric Wavelet-Based Dissimilarity Measure

In this paper, a wavelet-based multiscale asymmetric dissimilarity measure for shape matching is proposed. The wavelet transform is used to decompose the shape boundary into a multiscale representation. Given two shapes, a distance matrix is computed from the moment invariants of the wavelet coefficients at all the scale levels. The asymmetric dissimilarity is then calculated from the minimum values across each row on the distance matrix. The proposed asymmetric dissimilarity is a Hausdorff-like measure and is used for finding globally related shapes. The similarity paths obtained from the locations of the minimum distance values can be used to illustrate these relations.