Two-dimensional discrete shape matching and recognition

Abstract

We present a 2D matching method based on corresponding shape outlines. By working in discrete space, our study is done by using discrete operators and avoids interpolations and approximations. To encode shapes, we polygonalize their contours and we proceed by the extraction of intrinsic properties namely length, curvature and normal vectors. We optimize then a measure of similarity controlled by weight parameters over a dynamic programming process. The approach is not sensitive to sampling errors and affine transformations. We validate our approach on simple and complex forms, we made tests also to recognize shapes. The weight parameters could be interactively modified by an end-user to customize the matching. © Springer-Verlag Berlin Heidelberg 2006.