Planar shape representation based on multiwavelets

Abstract

A technique is presented to construct a muniscale representation of planar contours based on the multiwavelet transform (MWT). To generate this multiwavelet description, a partial 1-D discrete multiwavelet transform (DMWT) is applied to the vertical and horizontal components of a length-parametrized planar curve. This muniscale representation descomposes the curve into different levels of resolution and allows to reconstruct it to a desired degree of aproximation. A comparison between the multiwavelet, wavelet and the elliptic fourier transform (EFT) is presented. The results show that typical objects are well represented by a small number of multiwavelet coefficients allowing for a compact object shape representation.