Discrete wavelet descriptor for 2D shape

The aim of this paper is to propose a 2D shape recognition method under similitude transforms using the discrete dyadic wavelet transform with decimation (DWT). We propose four technics to fix the starting point necessary to obtain the same object representation in the framework of objects retrieval system. These representations are obtained by applying the DWT on the contour signature. We also propose a method to select a decomposition level of the DWT. These different solutions are assessed and compared using 1400 2D objects extracted from the MPEG7 database.