Learning of Form Models from Exemplars

Abstract

Model-based image recognition requires a general model of the object that should be detected. In many applications such models are not known a priori, but have to be learnt from examples. In this paper we describe our procedure for the acquisition and learning of general contour models. We developed a modified Procrustes algorithm for alignment and similarity calculation of shapes. Based on the calculated pair-wise similarity we learn groups of shapes. For each group we calculated prototypes. The set of prototypes will be used as models for the detection of object instances in new images. © Springer-Verlag 2004.