Wavelets for multiresolution shape recognition

Abstract

This paper describes a new method for 2-D shape recognition based on a multiresolution characterisation of the shape. From the Wavelet coefficients, features are extracted in order to perform a translation-rotation-scaling invariant recognition. Wavelets and multiresolution are exploited in order to reduce complexity of the matching task between the input image and the set of models. In the paper, motivations and performance of the algorithm are presented. Experimental results are also reported in several tests, including noise addition. The approach is quite general, and it could be extended to texture analysis, thus providing a unified paradigm for shape and texture recognition.